Article(id=1315743672671519655, tenantId=1146029695717560320, journalId=1315121910733963395, issueId=1315743622176288781, articleNumber=null, orderNo=null, doi=10.3788/gzxb20265502.0210003, pmid=null, cstr=32255.14.gzxb20265502.0210003, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1755705600000, receivedDateStr=2025-08-21, revisedDate=null, revisedDateStr=null, acceptedDate=1762099200000, acceptedDateStr=2025-11-03, onlineDate=1791627336900, onlineDateStr=2026-10-10, pubDate=1771948800000, pubDateStr=2026-02-25, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1791627336900, onlineIssueDateStr=2026-10-10, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1791627336900, creator=13701087609, updateTime=1791627336900, updator=13701087609, issue=Issue{id=1315743622176288781, tenantId=1146029695717560320, journalId=1315121910733963395, year='2026', volume='55', issue='2', pageStart='1', pageEnd='326', issueExtLink='null', onlineDate='null', pubDate='1771948800000', pubDateStr='2026-02-25', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1791627324861, creator='13701087609', updateTime=1791627390701, updator='13701087609', preIssue=null, nextIssue=null, articleTotal=null, ext={EN=IssueExt(id=1315743898396381488, tenantId=1146029695717560320, journalId=1315121910733963395, issueId=1315743622176288781, language=EN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1315743898396381489, tenantId=1146029695717560320, journalId=1315121910733963395, issueId=1315743622176288781, language=CN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null, downloadFileDto=null}, startPage=211, endPage=223, ext={EN=ArticleExt(id=1315743672935760809, articleId=1315743672671519655, tenantId=1146029695717560320, journalId=1315121910733963395, language=EN, title=Remote Sensing Image Object Detection Based on Adaptive Feature Weighting, columnId=1315743672860263336, journalTitle=Acta Photonica Sinica, columnName=Image Processing, runingTitle=null, highlight=null, articleAbstract=

The broad field of view inherent to remote sensing images offers significant advantages for large-scale monitoring tasks, enabling comprehensive surveillance over vast geographical areas with a single acquisition. However, this advantage is often offset by the presence of complex backgrounds, such as cloud cover, shadows, and varying terrain textures, which frequently introduce channel-level noise that interferes with target features. This interference significantly compromises the robustness and reliability of detection models, particularly in adverse weather conditions or densely cluttered environments. Additionally, the substantial variation in target sizes, ranging from small objects like vehicles and helicopters to large structures such as ports and bridges, poses a significant challenge. Detection models struggle to simultaneously capture high-resolution details necessary for identifying small targets and low-resolution semantic information essential for recognizing larger objects, leading to unbalanced detection performance across different scales.

To address the aforementioned challenges, this paper proposes a remote sensing image object detection algorithm based on adaptive feature weighting, built upon the Oriented R-CNN framework. The method introduces two core modules: the Adaptive Channel Weighting Module (ACWM) and the Adaptive Multi-Scale Fusion Pyramid (AMFP), aimed at tackling background interference and multi-scale target detection issues. The ACWM module is applied after the STAGE 2 and STAGE 3 phases of the feature extraction network to adaptively adjust channel weights, a critical step for enhancing feature representation. By analyzing statistical correlations between channels using the Correlation-Aware Covariance Weighting (CACW) method, it generates dynamic channel weights to enhance target-related features while suppressing background noise and redundant information, balancing the expression of different channels and thereby improving the feature extraction network's ability to extract features from rotated targets. Subsequently, the AMFP module processes the multi-scale features (C2, C3, C4, and C5) extracted by the feature extraction network, employing a hierarchical approach for feature integration. The Multi-scale Perception and Context Integration Module (MPCIM) fuses global semantic features from deep layers with edge and positional details from shallow layers, effectively addressing the issue of small targets being obscured. The Adaptive Layer Weighting Module (ALWM) optimizes the fusion quality of multi-scale features through adaptive layer-wise weighting, enhancing the model's capability to handle diverse target scales ranging from small vehicles to large ports by dynamically adjusting the contribution of each feature level, thus improving adaptability to various target scales. The optimized features (P2, P3, P4, and P5) are then fed into the Oriented RPN and Oriented RCNN head, which, based on the Oriented R-CNN framework, generates high-precision rotated bounding boxes to complete classification and localization tasks.

To validate the effectiveness of the proposed algorithm, extensive experiments were conducted on the DOTA1.0 and DIOR-R datasets. Ablation studies were performed to assess the contributions of the ACWM and AMFP modules. Results indicate that the ACWM module alone improves the mean Average Precision (mAP) by 0.97% on DOTA1.0 and 1.73% on DIOR-R, while the AMFP module contributes an additional 1.30% and 2.01% improvement, respectively. When both modules are combined, the mAP reaches 78.23% on DOTA1.0 and 68.17% on DIOR-R, representing overall enhancements of 2.36% and 3.87% compared to the baseline Oriented R-CNN model. Visualization comparisons further demonstrate the superiority of the proposed method, showing a significant reduction in missed detections and false positives for small targets, particularly in categories such as Small Vehicle, Swimming Pool, Helicopter, and Storage Tank. The method also exhibits improved localization accuracy and better contour fitting for rotated objects, especially in complex backgrounds with varying angles and dense distributions. The experimental results highlight the robustness and adaptability of the proposed algorithm across diverse remote sensing scenarios.

Compared to state-of-the-art methods such as Faster R-CNN, RoI Transformer, and ARC, the proposed approach achieves superior performance in handling small targets and complex orientations, as evidenced by its higher mAP and enhanced visual detection outcomes. However, limitations remain, including potential performance degradation in extremely cluttered scenes or with ultra-small targets, suggesting avenues for future optimization.

, authors=Yi LI, Fan LIU, Yating LIANG, authorsList=Yi LI, Fan LIU, Yating LIANG, authorCompany=null, correspAuthors=Fan LIU, authorNote=null, correspAuthorsNote=null, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, fund=null), CN=ArticleExt(id=1315743676014379965, articleId=1315743672671519655, tenantId=1146029695717560320, journalId=1315121910733963395, language=CN, title=基于自适应特征加权的遥感图像目标检测, columnId=1315743673036424106, journalTitle=光子学报, columnName=图像处理, runingTitle=null, highlight=null, articleAbstract=

针对光学遥感图像中背景干扰严重及目标尺度差异导致的检测精度低、漏检误检率高的问题,提出了一种基于自适应特征加权的遥感图像目标检测算法。在特征提取阶段中设计自适应通道加权模块,通过自适应通道加权增强目标相关特征、抑制背景噪声,并平衡多通道特征表达,有效缓解背景干扰的影响。同时,针对小目标和大目标的尺度差异,设计自适应多尺度融合金字塔,结合多尺度感知与上下文集成模块及自适应层间加权模块,动态优化多尺度特征融合,提升模型对小目标高分辨率细节捕捉和大目标低分辨率语义整合能力。在DOTA1.0和DIOR-R数据集上的实验表明,所提方法相较基线模型的平均精度分别提升2.36%和3.28%。可视化对比进一步验证了所提方法显著降低小目标的漏检和误检率。

, authors=李毅, 刘帆, 梁雅婷, authorsList=李毅, 刘帆, 梁雅婷, authorCompany=null, correspAuthors=刘帆, authorNote=

李毅,

, correspAuthorsNote=
刘帆,
, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=gSmqXIYlj+JpU3Tfrprm/g==, magXml=VoiGikolRp/c9BRCiFg2YQ==, pdfUrl=null, pdf=S+0qnNU+Cilos6OwWHKX8w==, pdfFileSize=7431387, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=T7bjjoXnLVpWQ1o//+dieA==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=hfX6O5WPG4HVW6JBlfyD0A==, mapNumber=null, fund=null)}, authors=[Author(id=1315743676425421762, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=liyi6630@163.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1315743676534473668, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676425421762, language=EN, stringName=Yi LI, firstName=Yi, middleName=null, lastName=LI, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1315743676601582533, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676425421762, language=CN, stringName=李毅, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1太原理工大学 人工智能学院, 晋中 030600, bio={"content":"

李毅,

"}, bioImg=null, bioContent=

李毅,

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1315743676324758462, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, xref=1, ext=[AuthorCompanyExt(id=1315743676333147071, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China), AuthorCompanyExt(id=1315743676345729984, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1太原理工大学 人工智能学院, 晋中 030600)])]), Author(id=1315743676685468615, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=liufan@tyut.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1315743676760966089, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676685468615, language=EN, stringName=Fan LIU, firstName=Fan, middleName=null, lastName=LIU, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1315743676832269258, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676685468615, language=CN, stringName=刘帆, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1太原理工大学 人工智能学院, 晋中 030600, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1315743676324758462, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, xref=1, ext=[AuthorCompanyExt(id=1315743676333147071, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China), AuthorCompanyExt(id=1315743676345729984, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1太原理工大学 人工智能学院, 晋中 030600)])]), Author(id=1315743676895183820, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1315743676966486990, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676895183820, language=EN, stringName=Yating LIANG, firstName=Yating, middleName=null, lastName=LIANG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1315743677054567375, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676895183820, language=CN, stringName=梁雅婷, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1太原理工大学 人工智能学院, 晋中 030600, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1315743676324758462, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, xref=1, ext=[AuthorCompanyExt(id=1315743676333147071, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China), AuthorCompanyExt(id=1315743676345729984, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1太原理工大学 人工智能学院, 晋中 030600)])])], keywords=[Keyword(id=1315743677163619280, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, orderNo=1, keyword=Deep learning), Keyword(id=1315743677234922449, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, orderNo=2, keyword=Object detection), Keyword(id=1315743677306225618, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, orderNo=3, keyword=Adaptive feature weighting), Keyword(id=1315743677373334483, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, orderNo=4, keyword=Optical remote sensing images), Keyword(id=1315743677444637652, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, orderNo=5, keyword=Multi-scale fusion), Keyword(id=1315743677503357909, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, orderNo=1, keyword=深度学习), Keyword(id=1315743677566272470, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, orderNo=2, keyword=目标检测), Keyword(id=1315743677633381335, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, orderNo=3, keyword=自适应特征加权), Keyword(id=1315743677692101592, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, orderNo=4, keyword=光学遥感图像), Keyword(id=1315743677759210457, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, orderNo=5, keyword=多尺度融合)], refs=[Reference(id=1315743681827685366, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2024, volume=null, issue=null, pageStart=27706, pageEnd=27716, url=null, language=null, rfNumber=1, rfOrder=0, authorNames=CAI Xinhao, LAI Qiuxia, WANG Yuwei, journalName=null, refType=null, unstructuredReference=CAI Xinhao, LAI Qiuxia, WANG Yuwei, et al. Poly kernel inception network for remote sensing detection[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024: 27706-27716., articleTitle=Poly kernel inception network for remote sensing detection, refAbstract=null), Reference(id=1315743681907377143, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2024, volume=36, issue=4, pageStart=5900, pageEnd=5920, url=null, language=null, rfNumber=2, rfOrder=1, authorNames=JIAO Licheng, WANG Mengjiao, LIU Xu, journalName=IEEE Transactions on Neural Networks and Learning Systems, refType=null, unstructuredReference=JIAO Licheng, WANG Mengjiao, LIU Xu, et al. Multiscale deep learning for detection and recognition: a comprehensive survey[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 36(4): 5900-5920., articleTitle=Multiscale deep learning for detection and recognition: a comprehensive survey, refAbstract=null), Reference(id=1315743681991263224, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2026, volume=55, issue=1, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=3, rfOrder=2, authorNames=LI Zongzhe, WANG Qicheng, YANG Weihua, journalName=Acta Photonica Sinica, refType=null, unstructuredReference=LI Zongzhe, WANG Qicheng, YANG Weihua, et al. Quantifying the effects of sensor noise and spatial resolution on small target detection in multispectral remote sensing imagery[J]. Acta Photonica Sinica, 2026, 55(1): 0110002., articleTitle=Quantifying the effects of sensor noise and spatial resolution on small target detection in multispectral remote sensing imagery, refAbstract=null), Reference(id=1315743682075149305, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2026, volume=55, issue=1, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=3, rfOrder=3, authorNames=李宗哲, 王琪程, 杨伟华, journalName=光子学报, refType=null, unstructuredReference=李宗哲, 王琪程, 杨伟华, 等. 光学遥感图像噪声和空间分辨率对小目标检测精度影响机制研究[J].光子学报, 2026, 55(1): 0110002., articleTitle=光学遥感图像噪声和空间分辨率对小目标检测精度影响机制研究, refAbstract=null), Reference(id=1315743682146452474, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2025, volume=158, issue=null, pageStart=110983, pageEnd=null, url=null, language=null, rfNumber=4, rfOrder=4, authorNames=GAO Tao, XIA Shilin, LIU Mengkun, journalName=Pattern Recognition, refType=null, unstructuredReference=GAO Tao, XIA Shilin, LIU Mengkun, et al. MSNet: multi-scale network for object detection in remote sensing images[J]. Pattern Recognition, 2025, 158: 110983., articleTitle=MSNet: multi-scale network for object detection in remote sensing images, refAbstract=null), Reference(id=1315743682226144251, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2025, volume=22, issue=null, pageStart=1, pageEnd=5, url=null, language=null, rfNumber=5, rfOrder=5, authorNames=ZHANG Yao, ZHAO Hongyang, LI Xingdong, journalName=IEEE Geoscience and Remote Sensing Letters, refType=null, unstructuredReference=ZHANG Yao, ZHAO Hongyang, LI Xingdong, et al. MERS-Net: a lightweight and efficient remote sensing image object detector[J]. IEEE Geoscience and Remote Sensing Letters, 2025, 22: 1-5., articleTitle=MERS-Net: a lightweight and efficient remote sensing image object detector, refAbstract=null), Reference(id=1315743682289058812, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2024, volume=53, issue=8, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=6, rfOrder=6, authorNames=ZENG Xiangjin, LIU Genghuan, CHEN Jianming, journalName=Acta Photonica Sinica, refType=null, unstructuredReference=ZENG Xiangjin, LIU Genghuan, CHEN Jianming, et al. A multi-scale hierarchical residual network-based method for tiny object detection in optical remote sensing images[J]. Acta Photonica Sinica, 2024, 53(8): 0810001., articleTitle=A multi-scale hierarchical residual network-based method for tiny object detection in optical remote sensing images, refAbstract=null), Reference(id=1315743682351973373, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2024, volume=53, issue=8, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=6, rfOrder=7, authorNames=曾祥津, 刘耿焕, 陈建明, journalName=光子学报, refType=null, unstructuredReference=曾祥津, 刘耿焕, 陈建明, 等. 基于多尺度分层残差网络的光学遥感图像微小目标检测[J].光子学报, 2024, 53(8): 0810001., articleTitle=基于多尺度分层残差网络的光学遥感图像微小目标检测, refAbstract=null), Reference(id=1315743682444248062, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=1, pageEnd=8, url=null, language=null, rfNumber=7, rfOrder=8, authorNames=CHO S J, JUN T J, OH B, journalName=null, refType=null, unstructuredReference=CHO S J, JUN T J, OH B, et al. DAPAS: Denoising autoencoder to prevent adversarial attack in semantic segmentation[C]. 2020 International Joint Conference on Neural Networks (IJCNN), 2020: 1-8., articleTitle=DAPAS: Denoising autoencoder to prevent adversarial attack in semantic segmentation, refAbstract=null), Reference(id=1315743682528134143, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2020, volume=124, issue=null, pageStart=117, pageEnd=129, url=null, language=null, rfNumber=8, rfOrder=9, authorNames=TIAN Chunwei, XU Yong, LI Zuoyong, journalName=Neural Networks, refType=null, unstructuredReference=TIAN Chunwei, XU Yong, LI Zuoyong, et al. Attention-guided CNN for image denoising[J]. Neural Networks, 2020, 124: 117-129., articleTitle=Attention-guided CNN for image denoising, refAbstract=null), Reference(id=1315743682591048704, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2022, volume=53, issue=3, pageStart=1954, pageEnd=1967, url=null, language=null, rfNumber=9, rfOrder=10, authorNames=LI Xiaotong, JIAO Licheng, ZHU Hao, journalName=IEEE Transactions on Cybernetics, refType=null, unstructuredReference=LI Xiaotong, JIAO Licheng, ZHU Hao, et al. A collaborative learning tracking network for remote sensing videos[J]. IEEE Transactions on Cybernetics, 2022, 53(3): 1954-1967., articleTitle=A collaborative learning tracking network for remote sensing videos, refAbstract=null), Reference(id=1315743682666545152, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=2117, pageEnd=2125, url=null, language=null, rfNumber=10, rfOrder=11, authorNames=LIN T Y, DOLLAR P, GIRSHICK R, journalName=null, refType=null, unstructuredReference=LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 2117-2125., articleTitle=Feature pyramid networks for object detection, refAbstract=null), Reference(id=1315743682742042625, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2022, volume=14, issue=15, pageStart=3735, pageEnd=null, url=null, language=null, rfNumber=11, rfOrder=12, authorNames=DONG Xiaohu, QIN Yao, GAO Yinghui, journalName=Remote Sensing, refType=null, unstructuredReference=DONG Xiaohu, QIN Yao, GAO Yinghui, et al. Attention-based multi-level feature fusion for object detection in remote sensing images[J]. Remote Sensing, 2022, 14(15): 3735., articleTitle=Attention-based multi-level feature fusion for object detection in remote sensing images, refAbstract=null), Reference(id=1315743682821734402, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2023, volume=19, issue=1, pageStart=1, pageEnd=22, url=null, language=null, rfNumber=12, rfOrder=13, authorNames=ZHEN Peining, WANG Shuqi, ZHANG Suming, journalName=ACM Transactions on Multimedia Computing, Communications and Applications, refType=null, unstructuredReference=ZHEN Peining, WANG Shuqi, ZHANG Suming, et al. Towards accurate oriented object detection in aerial images with adaptive multi-level feature fusion[J]. ACM Transactions on Multimedia Computing, Communications and Applications, 2023, 19(1): 1-22., articleTitle=Towards accurate oriented object detection in aerial images with adaptive multi-level feature fusion, refAbstract=null), Reference(id=1315743684461707267, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2024, volume=21, issue=null, pageStart=1, pageEnd=5, url=null, language=null, rfNumber=13, rfOrder=14, authorNames=GAO Yunxiao, WANG Yongcheng, ZHANG Yuxi, journalName=IEEE Geoscience and Remote Sensing Letters, refType=null, unstructuredReference=GAO Yunxiao, WANG Yongcheng, ZHANG Yuxi, et al. Feature super-resolution fusion with cross-scale distillation for small-object detection in optical remote sensing images[J]. IEEE Geoscience and Remote Sensing Letters, 2024, 21: 1-5., articleTitle=Feature super-resolution fusion with cross-scale distillation for small-object detection in optical remote sensing images, refAbstract=null), Reference(id=1315743684537204740, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=3520, pageEnd=3529, url=null, language=null, rfNumber=14, rfOrder=15, authorNames=XIE Xingxing, CHENG Gong, WANG Jiabao, journalName=null, refType=null, unstructuredReference=XIE Xingxing, CHENG Gong, WANG Jiabao, et al. Oriented R-CNN for object detection[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021: 3520-3529., articleTitle=Oriented R-CNN for object detection, refAbstract=null), Reference(id=1315743684600119301, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=3974, pageEnd=3983, url=null, language=null, rfNumber=15, rfOrder=16, authorNames=XIA Guisong, BAI Xiang, DING Jian, journalName=null, refType=null, unstructuredReference=XIA Guisong, BAI Xiang, DING Jian, et al. DOTA: a large-scale dataset for object detection in aerial images[C]. Proceedings of the IEEE conference on computer vision and pattern recognition, 2018: 3974-3983., articleTitle=DOTA: a large-scale dataset for object detection in aerial images, refAbstract=null), Reference(id=1315743684671422470, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2022, volume=60, issue=null, pageStart=1, pageEnd=11, url=null, language=null, rfNumber=16, rfOrder=17, authorNames=CHENG Gong, WANG Jiabao, LI Ke, journalName=IEEE Transactions on Geoscience and Remote Sensing, refType=null, unstructuredReference=CHENG Gong, WANG Jiabao, LI Ke, et al. Anchor-free oriented proposal generator for object detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-11., articleTitle=Anchor-free oriented proposal generator for object detection, refAbstract=null), Reference(id=1315743684751114247, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=16794, pageEnd=16805, url=null, language=null, rfNumber=17, rfOrder=18, authorNames=LI Yuxuan, HOU Qibin, ZHENG Zhaohui, journalName=null, refType=null, unstructuredReference=LI Yuxuan, HOU Qibin, ZHENG Zhaohui, et al. Large selective kernel network for remote sensing object detection[C]. Proceedings of the IEEE/CVF international conference on computer vision, 2023: 16794-16805., articleTitle=Large selective kernel network for remote sensing object detection, refAbstract=null), Reference(id=1315743684818223112, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=6589, pageEnd=6600, url=null, language=null, rfNumber=18, rfOrder=19, authorNames=PU Yifan, WANG Yiru, XIA Zhuofan, journalName=null, refType=null, unstructuredReference=PU Yifan, WANG Yiru, XIA Zhuofan, et al. Adaptive rotated convolution for rotated object detection[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023: 6589-6600., articleTitle=Adaptive rotated convolution for rotated object detection, refAbstract=null), Reference(id=1315743684893720585, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=2786, pageEnd=2795, url=null, language=null, rfNumber=19, rfOrder=20, authorNames=HAN Jiaming, DING Jian, XUE Nan, journalName=null, refType=null, unstructuredReference=HAN Jiaming, DING Jian, XUE Nan, et al. ReDet: a rotation-equivariant detector for aerial object detection[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 2786-2795., articleTitle=ReDet: a rotation-equivariant detector for aerial object detection, refAbstract=null), Reference(id=1315743684981800970, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2009, volume=null, issue=null, pageStart=248, pageEnd=255, url=null, language=null, rfNumber=20, rfOrder=21, authorNames=DENG Jia, DONG Wei, SOCHER R, journalName=null, refType=null, unstructuredReference=DENG Jia, DONG Wei, SOCHER R, et al. Imagenet: a large-scale hierarchical image database[C]. 2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009: 248-255., articleTitle=Imagenet: a large-scale hierarchical image database, refAbstract=null), Reference(id=1315743685057298443, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=21, rfOrder=22, authorNames=MMPRETRAIN CONTRIBUTORS, journalName=null, refType=null, unstructuredReference=MMPRETRAIN CONTRIBUTORS. OpenMMLab's pre-training toolbox and benchmark[EB/OL]. https://github.com/open-mmlab/mmpretrain, 2023., articleTitle=OpenMMLab's pre-training toolbox and benchmark, refAbstract=null), Reference(id=1315743685120213004, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=22, rfOrder=23, authorNames=LOSHCHILOV I, HUTTER F, journalName=arXiv preprint, refType=null, unstructuredReference=LOSHCHILOV I, HUTTER F. Decoupled weight decay regularization[EB/OL]. arXiv preprint arXiv:2017., articleTitle=Decoupled weight decay regularization, refAbstract=null), Reference(id=1315743685199904781, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=7331, pageEnd=7334, url=null, language=null, rfNumber=23, rfOrder=24, authorNames=ZHOU Yue, YANG Xue, ZHANG Gefan, journalName=null, refType=null, unstructuredReference=ZHOU Yue, YANG Xue, ZHANG Gefan, et al. Mmrotate: a rotated object detection benchmark using pytorch[C]. Proceedings of the 30th ACM International Conference on Multimedia, 2022: 7331-7334., articleTitle=Mmrotate: a rotated object detection benchmark using pytorch, refAbstract=null), Reference(id=1315743685275402254, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=2980, pageEnd=2988, url=null, language=null, rfNumber=24, rfOrder=25, authorNames=LIN T Y, GOYAL P, GIRSHICK R, journalName=null, refType=null, unstructuredReference=LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]. Proceedings of the IEEE International Conference on Computer Vision, 2017: 2980-2988., articleTitle=Focal loss for dense object detection, refAbstract=null), Reference(id=1315743685409619983, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2016, volume=39, issue=6, pageStart=1137, pageEnd=1149, url=null, language=null, rfNumber=25, rfOrder=26, authorNames=REN S Q, HE K M, GIRSHICK R, journalName=IEEE transactions on pattern analysis and machine intelligence, refType=null, unstructuredReference=REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2016, 39(6): 1137-1149., articleTitle=Faster R-CNN: towards real-time object detection with region proposal networks, refAbstract=null), Reference(id=1315743685501894672, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=2849, pageEnd=2858, url=null, language=null, rfNumber=26, rfOrder=27, authorNames=DING Jian, XUE Nan, LONG Yang, journalName=null, refType=null, unstructuredReference=DING Jian, XUE Nan, LONG Yang, et al. Learning RoI transformer for oriented object detection in aerial images[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 2849-2858., articleTitle=Learning RoI transformer for oriented object detection in aerial images, refAbstract=null), Reference(id=1315743685577392145, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=11207, pageEnd=11216, url=null, language=null, rfNumber=27, rfOrder=28, authorNames=PAN Xingjia, REN Yuqiang, SHENG Kekai, journalName=null, refType=null, unstructuredReference=PAN Xingjia, REN Yuqiang, SHENG Kekai, et al. Dynamic refinement network for oriented and densely packed object detection[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 11207-11216., articleTitle=Dynamic refinement network for oriented and densely packed object detection, refAbstract=null), Reference(id=1315743685736775698, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2021, volume=60, issue=null, pageStart=1, pageEnd=11, url=null, language=null, rfNumber=28, rfOrder=29, authorNames=HAN Jiaming, DING Jian, LI Jie, journalName=IEEE Transactions on Geoscience and Remote Sensing, refType=null, unstructuredReference=HAN Jiaming, DING Jian, LI Jie, et al. Align deep features for oriented object detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60: 1-11., articleTitle=Align deep features for oriented object detection, refAbstract=null), Reference(id=1315743685812273171, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2024, volume=62, issue=null, pageStart=1, pageEnd=15, url=null, language=null, rfNumber=29, rfOrder=30, authorNames=ZENG Ying, CHEN Yushi, YANG Xue, journalName=IEEE Transactions on Geoscience and Remote Sensing, refType=null, unstructuredReference=ZENG Ying, CHEN Yushi, YANG Xue, et al. ARS-DETR: aspect ratio-sensitive detection transformer for aerial oriented object detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 1-15., articleTitle=ARS-DETR: aspect ratio-sensitive detection transformer for aerial oriented object detection, refAbstract=null), Reference(id=1315743685875187732, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2024, volume=62, issue=null, pageStart=1, pageEnd=10, url=null, language=null, rfNumber=30, rfOrder=31, authorNames=XIE Xingxing, CHENG Gong, RAO Chaofan, journalName=IEEE Transactions on Geoscience and Remote Sensing, refType=null, unstructuredReference=XIE Xingxing, CHENG Gong, RAO Chaofan, et al. Oriented object detection via contextual dependence mining and penalty-incentive allocation[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 1-10., articleTitle=Oriented object detection via contextual dependence mining and penalty-incentive allocation, refAbstract=null), Reference(id=1315743685938102293, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=6589, pageEnd=6600, url=null, language=null, rfNumber=31, rfOrder=32, authorNames=PU Yifan, WANG Yiru, XIA Zhuofan, journalName=null, refType=null, unstructuredReference=PU Yifan, WANG Yiru, XIA Zhuofan, et al. Adaptive rotated convolution for rotated object detection[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023: 6589-6600., articleTitle=Adaptive rotated convolution for rotated object detection, refAbstract=null), Reference(id=1315743686021988374, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2023, volume=62, issue=null, pageStart=1, pageEnd=14, url=null, language=null, rfNumber=32, rfOrder=33, authorNames=GONG Maoguo, ZHAO Hongyu, WU Yue, journalName=IEEE Transactions on Geoscience and Remote Sensing, refType=null, unstructuredReference=GONG Maoguo, ZHAO Hongyu, WU Yue, et al. Dual appearance-aware enhancement for oriented object detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 62: 1-14., articleTitle=Dual appearance-aware enhancement for oriented object detection, refAbstract=null), Reference(id=1315743686105874455, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, doi=null, pmid=null, pmcid=null, year=2022, volume=60, issue=null, pageStart=1, pageEnd=11, url=null, language=null, rfNumber=33, rfOrder=34, authorNames=CHENG Gong, YAO Yanqing, LI Shengyang, journalName=IEEE Transactions on Geoscience and Remote Sensing, refType=null, unstructuredReference=CHENG Gong, YAO Yanqing, LI Shengyang, et al. Dual-aligned oriented detector[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-11., articleTitle=Dual-aligned oriented detector, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1315743676324758462, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, xref=1, ext=[AuthorCompanyExt(id=1315743676333147071, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China), AuthorCompanyExt(id=1315743676345729984, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1太原理工大学 人工智能学院, 晋中 030600)])], figs=[ArticleFig(id=1315743677897622490, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Fig.1, caption=The overall network structure, figureFileSmall=ZrNFG3SP+OYQMskbqS3y2g==, figureFileBig=T7bjjoXnLVpWQ1o//+dieA==, tableContent=null), ArticleFig(id=1315743677981508571, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=图1, caption=整体网络结构, figureFileSmall=ZrNFG3SP+OYQMskbqS3y2g==, figureFileBig=T7bjjoXnLVpWQ1o//+dieA==, tableContent=null), ArticleFig(id=1315743678170252252, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Fig.2, caption=Adaptive channel weighting module, figureFileSmall=dacd8OSI3wD3o43ti2xPOg==, figureFileBig=0BQNaVFJvS+I5tTtB7nYDw==, tableContent=null), ArticleFig(id=1315743678241555421, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=图2, caption=自适应通道加权模块, figureFileSmall=dacd8OSI3wD3o43ti2xPOg==, figureFileBig=0BQNaVFJvS+I5tTtB7nYDw==, tableContent=null), ArticleFig(id=1315743678329635806, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Fig.3, caption=Correlation-aware covariance weighting, figureFileSmall=zWSYPedgLA6zVdZBscJVIg==, figureFileBig=ie0NLTWgYuTv8YvDkNVkIg==, tableContent=null), ArticleFig(id=1315743678396744671, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=图3, caption=相关性感知协方差加权, figureFileSmall=zWSYPedgLA6zVdZBscJVIg==, figureFileBig=ie0NLTWgYuTv8YvDkNVkIg==, tableContent=null), ArticleFig(id=1315743680045106144, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Fig.4, caption=Multi-scale perception and context integration module, figureFileSmall=bELz6gM3LDZMzY3ihQMaJw==, figureFileBig=fwCvYKLyyELZscmpqDB9eA==, tableContent=null), ArticleFig(id=1315743680112215009, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=图4, caption=多尺度感知与上下文集成模块, figureFileSmall=bELz6gM3LDZMzY3ihQMaJw==, figureFileBig=fwCvYKLyyELZscmpqDB9eA==, tableContent=null), ArticleFig(id=1315743680179323874, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Fig.5, caption=Adaptive layer weighting module, figureFileSmall=PvoW+awrBMN7bImkiklMyA==, figureFileBig=t3LvydroyIx2rYArf4ZjAg==, tableContent=null), ArticleFig(id=1315743680259015651, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=图5, caption=自适应层间加权模块, figureFileSmall=PvoW+awrBMN7bImkiklMyA==, figureFileBig=t3LvydroyIx2rYArf4ZjAg==, tableContent=null), ArticleFig(id=1315743680326124516, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Fig.6, caption=Comparison of detection results between proposed method and baseline method. (a)(f) Dense vehicle; (b)(g) Airport area target; (c)(h) Small target in water area; (d)(i) Sports field and surrounding area; (e)(j) Dock ship, figureFileSmall=FTJA526ciIpRWMmc3Jw7Aw==, figureFileBig=nrt2dZ6VDoawIwm7rAhq+Q==, tableContent=null), ArticleFig(id=1315743680410010597, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=图6, caption=所提方法与基线方法检测结果对比。(a)(f) 密集车辆;(b)(g) 机场区域目标;(c)(h) 水域小型目标;(d)(i) 运动场地及周边;(e)(j) 码头船只, figureFileSmall=FTJA526ciIpRWMmc3Jw7Aw==, figureFileBig=nrt2dZ6VDoawIwm7rAhq+Q==, tableContent=null), ArticleFig(id=1315743680489702374, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Fig.7, caption=PR curves of the proposed method and baseline method in different categories. (a) The PR curve of Airplane; (b) The PR curve of Ship, figureFileSmall=CpAVUa2e9oXmrB1tkcA42w==, figureFileBig=hSfQkEaanYkTgHOIZIYRSA==, tableContent=null), ArticleFig(id=1315743680565199847, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=图7, caption=所提方法与基线方法在不同类别上的PR曲线。(a) Airplane的PR曲线;(b) Ship的PR曲线, figureFileSmall=CpAVUa2e9oXmrB1tkcA42w==, figureFileBig=hSfQkEaanYkTgHOIZIYRSA==, tableContent=null), ArticleFig(id=1315743680636503016, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Fig.8, caption=Comparison of detection results of different algorithms on the DOTA1.0 dataset. (a) RetinaNet-O; (b) Faster R-CNN-O; (c) Roi Transformer; (d) ARS-DETR; (e) Oriented R-CNN; (f) The proposed method, figureFileSmall=R7k1o/iYqiscj3q9oAeAVw==, figureFileBig=wnchhtwptaPRTE6knjespw==, tableContent=null), ArticleFig(id=1315743680724583401, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=图8, caption=不同算法在DOTA1.0数据集上的检测结果对比。(a) RetinaNet-O; (b) Faster R-CNN-O; (c) Roi Transformer; (d) ARS-DETR;(e) Oriented R-CNN;(f)本文方法, figureFileSmall=R7k1o/iYqiscj3q9oAeAVw==, figureFileBig=wnchhtwptaPRTE6knjespw==, tableContent=null), ArticleFig(id=1315743680795886570, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Fig.9, caption=The detection results of the proposed method on the DIOR-R dataset, figureFileSmall=vde0x6VVElCxhnyPg1TzPQ==, figureFileBig=fc6bTP33uuAv0t0koeCXWw==, tableContent=null), ArticleFig(id=1315743680854606827, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=图9, caption=所提方法在DIOR-R数据集上的检测结果, figureFileSmall=vde0x6VVElCxhnyPg1TzPQ==, figureFileBig=fc6bTP33uuAv0t0koeCXWw==, tableContent=null), ArticleFig(id=1315743680934298604, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Table 1, caption=

Ablation experiment on the DOTA1.0 dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsACWMAMFPmAP↑
Oriented R-CNN××75.87
Oriented R-CNN+ACWM√×76.84
Oriented R-CNN+AMFP×√77.17
Oriented R-CNN+ACWM+AMFP√√78.23
), ArticleFig(id=1315743681018184685, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=表1, caption=

在DOTA1.0数据集上的消融实验

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsACWMAMFPmAP↑
Oriented R-CNN××75.87
Oriented R-CNN+ACWM√×76.84
Oriented R-CNN+AMFP×√77.17
Oriented R-CNN+ACWM+AMFP√√78.23
), ArticleFig(id=1315743681106265070, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Table 2, caption=

Ablation experiment on the DIOR-R dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsACWMAMFPmAP↑mAP(CM)mAP(VM)
Oriented R-CNN××64.3072.6046.80
Oriented R-CNN+ACWM√×66.0374.7649.78
Oriented R-CNN+AMFP×√66.3175.2150.05
Oriented R-CNN+ACWM+AMFP√√67.5876.1951.86
), ArticleFig(id=1315743681190151151, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=表2, caption=

在DIOR-R数据集上的消融实验

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsACWMAMFPmAP↑mAP(CM)mAP(VM)
Oriented R-CNN××64.3072.6046.80
Oriented R-CNN+ACWM√×66.0374.7649.78
Oriented R-CNN+AMFP×√66.3175.2150.05
Oriented R-CNN+ACWM+AMFP√√67.5876.1951.86
), ArticleFig(id=1315743681265648624, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Table 3, caption=

Comparison of detection accuracy of different algorithms on DOTA1.0

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsBackboneLarge ObjectMedium ObjectSmall ObjectmAP↑
BDSBFBRGTFBCHAPLTCLVSTSPRASHSVHC
RetinaNet-O[24]ResNet-5077.6254.7541.8158.1782.1862.5788.6790.2971.6474.3269.5760.6079.1174.5860.6468.43
Faster R-CNN-O[25]ResNet-5073.0648.5944.8659.0978.9462.1888.4490.8471.4983.9064.9162.9577.1173.2556.1869.05
S2ANet[28]ResNet-5082.8460.3648.3771.1184.9065.2689.1190.8378.3985.6469.1362.6087.2578.1157.9474.12
ARS-DETR[29]ResNet-5075.5660.2848.3269.2077.3174.8886.9790.5077.9482.8671.7664.5887.6977.9266.6274.16
DFDet[30]ResNet-5079.2560.0748.4070.0083.1368.2788.9290.9078.8583.9876.7866.4987.2180.2258.1174.71
Oriented R-CNN[14]ResNet-5082.1263.9754.7870.8687.5074.9489.4690.9083.0084.6868.8467.6988.2078.9352.2875.87
RoI Transformer[26]ResNet-10178.5258.3943.4475.9277.2762.8388.6490.7473.6881.4658.9353.5483.5968.8147.6769.56
DRN[27]Hourglass-5280.2250.1243.5263.3583.8567.6288.9190.1470.6984.1168.6058.4184.9473.4852.5070.70
ARC[31]ARC-R10182.4863.6355.3373.8886.4474.2989.4090.9084.0584.8371.9170.3288.0679.3765.4377.35
OursResNet-5082.9268.2054.6377.5886.9576.3589.5390.9584.8784.9372.0170.3988.2379.3766.6878.23
), ArticleFig(id=1315743681353729009, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=表3, caption=

不同算法在DOTA1.0数据集上检测精度对比

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsBackboneLarge ObjectMedium ObjectSmall ObjectmAP↑
BDSBFBRGTFBCHAPLTCLVSTSPRASHSVHC
RetinaNet-O[24]ResNet-5077.6254.7541.8158.1782.1862.5788.6790.2971.6474.3269.5760.6079.1174.5860.6468.43
Faster R-CNN-O[25]ResNet-5073.0648.5944.8659.0978.9462.1888.4490.8471.4983.9064.9162.9577.1173.2556.1869.05
S2ANet[28]ResNet-5082.8460.3648.3771.1184.9065.2689.1190.8378.3985.6469.1362.6087.2578.1157.9474.12
ARS-DETR[29]ResNet-5075.5660.2848.3269.2077.3174.8886.9790.5077.9482.8671.7664.5887.6977.9266.6274.16
DFDet[30]ResNet-5079.2560.0748.4070.0083.1368.2788.9290.9078.8583.9876.7866.4987.2180.2258.1174.71
Oriented R-CNN[14]ResNet-5082.1263.9754.7870.8687.5074.9489.4690.9083.0084.6868.8467.6988.2078.9352.2875.87
RoI Transformer[26]ResNet-10178.5258.3943.4475.9277.2762.8388.6490.7473.6881.4658.9353.5483.5968.8147.6769.56
DRN[27]Hourglass-5280.2250.1243.5263.3583.8567.6288.9190.1470.6984.1168.6058.4184.9473.4852.5070.70
ARC[31]ARC-R10182.4863.6355.3373.8886.4474.2989.4090.9084.0584.8371.9170.3288.0679.3765.4377.35
OursResNet-5082.9268.2054.6377.5886.9576.3589.5390.9584.8784.9372.0170.3988.2379.3766.6878.23
), ArticleFig(id=1315743681429226482, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Table 4, caption=

Performance comparison of different algorithms on DOTA 1.0

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsParams↓FLOPs↓mAP↑
RetinaNet-O[24]38.8 M203 G68.43
Faster R-CNN-O[25]41.1 M211 G69.05
S2ANet[28]38.6 M198 G74.12
ARS-DETR[29]41.8 M208 G74.16
DFDet[30]34.4 M186 G74.71
Oriented R-CNN[14]41.4 M206 G75.87
RoI Transformer[26]55.1 M225 G69.56
DRN[27]--70.70
ARC[31]74.4 M232 G77.35
Ours41.6 M210 G78.23
), ArticleFig(id=1315743681517306867, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=表4, caption=

不同算法在DOTA 1.0数据集上的性能对比

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsParams↓FLOPs↓mAP↑
RetinaNet-O[24]38.8 M203 G68.43
Faster R-CNN-O[25]41.1 M211 G69.05
S2ANet[28]38.6 M198 G74.12
ARS-DETR[29]41.8 M208 G74.16
DFDet[30]34.4 M186 G74.71
Oriented R-CNN[14]41.4 M206 G75.87
RoI Transformer[26]55.1 M225 G69.56
DRN[27]--70.70
ARC[31]74.4 M232 G77.35
Ours41.6 M210 G78.23
), ArticleFig(id=1315743681605387252, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=EN, label=Table 5, caption=

Comparison of detection accuracy of different algorithms on DIOR-R

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsLarge ObjectMedium ObjectSmall ObjectmAP↑
GFATSDHBBGGCDMOPBFSTBCWMTSTCALEACMESVMSP
Ours83.2946.1979.8145.5945.0176.5834.5358.9178.8169.8888.0465.9159.4282.6971.7573.2776.1978.8351.8684.0967.58
RetinaNet-O[24]79.5428.5267.5732.1423.9869.2519.9444.8773.5761.0981.1760.2447.3381.4661.4972.3972.5458.2038.0177.7157.55
Faster
R-CNN-O[25]
79.2426.8064.3434.9534.2066.6318.9548.7971.7271.2180.9165.2147.3181.4462.7966.4572.5765.7550.4681.1459.54
ARS-DETR[29]78.6254.1770.0936.3341.1376.1034.8955.4176.4472.2381.6570.2861.5281,1468.0073.0775.6681.8248.5884.5566.12
DFDet[30]79.6238.8373.1138.2634.1175.5626.2652.7677.4168.2781.3663.3552.2381.3861.9276.0674.9762.3144.1180.4062.11
Oriented
R-CNN[14]
82.4039.3074.2043.6043.1068.6029.5057.6079.5062.6086.2066.0054.8081.4071.1066.7072.6079.3046.8081.3064.30
RoI
Transformer[26]
82.7437.8878.2347.7140.6868.9626.8655.6171.7870.2687.5365.5254.8681.6163.3478.7172.6068.0943.2781.2163.87
ASDNet[32]80.6230.3374.6241.9236.8466.8725.0852.8871.4171.0081.4362.9154.2881.4862.8874.9772.5160.6547.4981.2161.57
DODet[33]75.4743.3574.0448.0043.1274.1533.3259.3172.1171.5681.3266.4056.7681.5263.4078.7772.5970.8450.5785.4165.10
), ArticleFig(id=1315743681701856245, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, language=CN, label=表5, caption=

不同算法在DIOR-R数据集上检测精度对比

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsLarge ObjectMedium ObjectSmall ObjectmAP↑
GFATSDHBBGGCDMOPBFSTBCWMTSTCALEACMESVMSP
Ours83.2946.1979.8145.5945.0176.5834.5358.9178.8169.8888.0465.9159.4282.6971.7573.2776.1978.8351.8684.0967.58
RetinaNet-O[24]79.5428.5267.5732.1423.9869.2519.9444.8773.5761.0981.1760.2447.3381.4661.4972.3972.5458.2038.0177.7157.55
Faster
R-CNN-O[25]
79.2426.8064.3434.9534.2066.6318.9548.7971.7271.2180.9165.2147.3181.4462.7966.4572.5765.7550.4681.1459.54
ARS-DETR[29]78.6254.1770.0936.3341.1376.1034.8955.4176.4472.2381.6570.2861.5281,1468.0073.0775.6681.8248.5884.5566.12
DFDet[30]79.6238.8373.1138.2634.1175.5626.2652.7677.4168.2781.3663.3552.2381.3861.9276.0674.9762.3144.1180.4062.11
Oriented
R-CNN[14]
82.4039.3074.2043.6043.1068.6029.5057.6079.5062.6086.2066.0054.8081.4071.1066.7072.6079.3046.8081.3064.30
RoI
Transformer[26]
82.7437.8878.2347.7140.6868.9626.8655.6171.7870.2687.5365.5254.8681.6163.3478.7172.6068.0943.2781.2163.87
ASDNet[32]80.6230.3374.6241.9236.8466.8725.0852.8871.4171.0081.4362.9154.2881.4862.8874.9772.5160.6547.4981.2161.57
DODet[33]75.4743.3574.0448.0043.1274.1533.3259.3172.1171.5681.3266.4056.7681.5263.4078.7772.5970.8450.5785.4165.10
)], attaches=null, journal=Journal(id=1315120956596912254, delFlag=0, nameCn=光子学报, nameEn=Acta Photonica Sinica, nameHistory1=null, nameHistory2=null, issn=1004-4213, eissn=null, cn=61-1235/O4, coden=null, periodic=0, language=CN, oaType=null, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=V4CYNtLUy6tkm8zxiX3hkw==, journalPrice=null, startedYear=null, abbrevIsoEn=Acta Photonica Sinica, journalRemark=null, publicationField=null, createdTime=1791478869821, updatedTime=1791802324400, createdBy=18614031015, updatedBy=13701087609, firstLetterCn=G, firstLetterEn=G, subjectCode=Engineering, subjectName=null, subjectCodeEn=Engineering, subjectNameEn=null, picCn=V4CYNtLUy6tkm8zxiX3hkw==, picEn=kSVLl0/HjBthS+BEgcahLw==, jcr=null, cjcr=null, exts=[JournalExt(id=1316477625225298568, language=CN, name=光子学报, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1791802324825, updatedTime=1791802324825, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://gzxb.portal.founderss.cn, submissionEditorUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-portal&redirect_uri=https%3A%2F%2Fjournal.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=69eadfdc-ccc1-45af-b, submissionReviewUrl=https://gzxb.portal.founderss.cn, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1316477625288213129, language=EN, name=Acta Photonica Sinica, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1791802324840, updatedTime=1791802324840, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://gzxb.portal.founderss.cn, submissionEditorUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-portal&redirect_uri=https%3A%2F%2Fjournal.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=69eadfdc-ccc1-45af-b, submissionReviewUrl=https://gzxb.portal.founderss.cn, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1315121910733963395, websiteList=[Website(id=1315346106637210224, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1315121910733963395, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/gzxb/CN, language=CN, createTime=1791532549771, createBy=18614031015, updateTime=1791533256542, updateBy=18614031015, name=光子学报-中文, tplId=1146099689490845704, title=光子学报, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1315349214071210463, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346106637210224, code=articleTextType, value=kx, createTime=1791533290641, updateTime=1791533290641, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349213987324380, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346106637210224, code=banner, value=null, createTime=1791533290621, updateTime=1791533290621, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349214134125026, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346106637210224, code=grayFlag, value=0, createTime=1791533290656, updateTime=1791533290656, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349213966352859, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346106637210224, code=logo, value=https://castjournals.cast.org.cn/joweb/gzxb/CN/file/pic?fileId=wQBjM6S+LBsZjnXGl57u5g==, createTime=1791533290616, updateTime=1791533290616, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349214171873764, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346106637210224, code=minRunFlag, value=0, createTime=1791533290665, updateTime=1791533290665, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349214054433247, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346106637210224, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/gzxb/CN/file/pic, createTime=1791533290637, updateTime=1791533290637, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349214155096547, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346106637210224, code=silenceFlag, value=0, createTime=1791533290661, updateTime=1791533290661, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349214020878813, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346106637210224, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1791533290629, updateTime=1791533290629, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349214083793376, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346106637210224, code=themeColor, value=null, createTime=1791533290644, updateTime=1791533290644, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349214104764897, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346106637210224, code=themeStyle, value=null, createTime=1791533290649, updateTime=1791533290649, creator=18614031015, updator=18614031015)]), Website(id=1315346107014697606, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1315121910733963395, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/gzxb/EN, language=EN, createTime=1791532549861, createBy=18614031015, updateTime=1791533251338, updateBy=18614031015, name=光子学报-英文, tplId=1146101810881728533, title=Acta Photonica Sinica, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1315349179224932817, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346107014697606, code=articleTextType, value=kx, createTime=1791533282333, updateTime=1791533282333, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349179187184078, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346107014697606, code=banner, value=null, createTime=1791533282324, updateTime=1791533282324, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349179271070164, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346107014697606, code=grayFlag, value=0, createTime=1791533282344, updateTime=1791533282344, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349179162018253, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346107014697606, code=logo, value=https://castjournals.cast.org.cn/joweb/gzxb/EN/file/pic?fileId=wQBjM6S+LBsZjnXGl57u5g==, createTime=1791533282318, updateTime=1791533282318, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349179292041686, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346107014697606, code=minRunFlag, value=0, createTime=1791533282349, updateTime=1791533282349, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349179208155600, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346107014697606, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/gzxb/EN/file/pic, createTime=1791533282329, updateTime=1791533282329, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349179279458773, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346107014697606, code=silenceFlag, value=0, createTime=1791533282346, updateTime=1791533282346, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349179199766991, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346107014697606, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1791533282327, updateTime=1791533282327, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349179233321426, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346107014697606, code=themeColor, value=null, createTime=1791533282335, updateTime=1791533282335, creator=18614031015, updator=18614031015), WebsiteProps(id=1315349179258487251, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1315346107014697606, code=themeStyle, value=null, createTime=1791533282341, updateTime=1791533282341, creator=18614031015, updator=18614031015)])], journalTitle=光子学报, weixinUrl=null, journalUrl=https://www.photon.ac.cn/, iacademicId=null, status=1, seqNo=null, journalTitleEn=Acta Photonica Sinica, journalPhotoCn=V4CYNtLUy6tkm8zxiX3hkw==, journalPhotoEn=kSVLl0/HjBthS+BEgcahLw==, journalFirstLetter=G, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=, provinceCode=null, provinceName=null, collectFlag=false, interPubPlatform=, interPubPlatformUrl=null), detailUrlCn=https://castjournals.cast.org.cn/joweb/gzxb/CN/10.3788/gzxb20265502.0210003, detailUrlEn=https://castjournals.cast.org.cn/joweb/gzxb/EN/10.3788/gzxb20265502.0210003, pdfUrlCn=https://castjournals.cast.org.cn/joweb/gzxb/CN/PDF/10.3788/gzxb20265502.0210003, pdfUrlEn=https://castjournals.cast.org.cn/joweb/gzxb/EN/PDF/10.3788/gzxb20265502.0210003, aliStartDate=0, aliEndDate=0, collectionFlag=false, citedCount=null, citedUrl=null, previewStatus=0, delFlag=0, hasFullText=1, orderTime=1771948800000, fullTextJson=null, articleText=null, reference=null)
收藏切换
基于自适应特征加权的遥感图像目标检测
收藏切换
PDF下载
李毅 , 刘帆 , 梁雅婷
光子学报 | 图像处理 2026,55(2): 211-223
收起
收藏切换
光子学报 |图像处理 2026 , 55 (2) : 211 -223
基于自适应特征加权的遥感图像目标检测
全屏
[Author(id=1315743676425421762, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=liyi6630@163.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1315743676534473668, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676425421762, language=EN, stringName=Yi LI, firstName=Yi, middleName=null, lastName=LI, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1315743676601582533, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676425421762, language=CN, stringName=李毅, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1太原理工大学 人工智能学院, 晋中 030600, bio={"content":"

李毅,

"}, bioImg=null, bioContent=

李毅,

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1315743676324758462, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, xref=1, ext=[AuthorCompanyExt(id=1315743676333147071, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China), AuthorCompanyExt(id=1315743676345729984, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1太原理工大学 人工智能学院, 晋中 030600)])]), Author(id=1315743676685468615, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=liufan@tyut.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1315743676760966089, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676685468615, language=EN, stringName=Fan LIU, firstName=Fan, middleName=null, lastName=LIU, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1315743676832269258, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676685468615, language=CN, stringName=刘帆, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1太原理工大学 人工智能学院, 晋中 030600, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1315743676324758462, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, xref=1, ext=[AuthorCompanyExt(id=1315743676333147071, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China), AuthorCompanyExt(id=1315743676345729984, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1太原理工大学 人工智能学院, 晋中 030600)])]), Author(id=1315743676895183820, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1315743676966486990, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676895183820, language=EN, stringName=Yating LIANG, firstName=Yating, middleName=null, lastName=LIANG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1315743677054567375, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, authorId=1315743676895183820, language=CN, stringName=梁雅婷, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1太原理工大学 人工智能学院, 晋中 030600, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1315743676324758462, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, xref=1, ext=[AuthorCompanyExt(id=1315743676333147071, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China), AuthorCompanyExt(id=1315743676345729984, tenantId=1146029695717560320, journalId=1315121910733963395, articleId=1315743672671519655, companyId=1315743676324758462, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1太原理工大学 人工智能学院, 晋中 030600)])])]
李毅 , 刘帆 , 梁雅婷
作者信息
  • 1太原理工大学 人工智能学院, 晋中 030600
通讯作者:
作者简介:
Remote Sensing Image Object Detection Based on Adaptive Feature Weighting
Yi LI , Fan LIU , Yating LIANG
Affiliations
  • 1College of Artificial Intelligence, Taiyuan University of Technology, Jinzhong 030600, China
出版时间: 2026-02-25 doi: 10.3788/gzxb20265502.0210003
文章导航
收藏切换

针对光学遥感图像中背景干扰严重及目标尺度差异导致的检测精度低、漏检误检率高的问题,提出了一种基于自适应特征加权的遥感图像目标检测算法。在特征提取阶段中设计自适应通道加权模块,通过自适应通道加权增强目标相关特征、抑制背景噪声,并平衡多通道特征表达,有效缓解背景干扰的影响。同时,针对小目标和大目标的尺度差异,设计自适应多尺度融合金字塔,结合多尺度感知与上下文集成模块及自适应层间加权模块,动态优化多尺度特征融合,提升模型对小目标高分辨率细节捕捉和大目标低分辨率语义整合能力。在DOTA1.0和DIOR-R数据集上的实验表明,所提方法相较基线模型的平均精度分别提升2.36%和3.28%。可视化对比进一步验证了所提方法显著降低小目标的漏检和误检率。

深度学习  /  目标检测  /  自适应特征加权  /  光学遥感图像  /  多尺度融合

The broad field of view inherent to remote sensing images offers significant advantages for large-scale monitoring tasks, enabling comprehensive surveillance over vast geographical areas with a single acquisition. However, this advantage is often offset by the presence of complex backgrounds, such as cloud cover, shadows, and varying terrain textures, which frequently introduce channel-level noise that interferes with target features. This interference significantly compromises the robustness and reliability of detection models, particularly in adverse weather conditions or densely cluttered environments. Additionally, the substantial variation in target sizes, ranging from small objects like vehicles and helicopters to large structures such as ports and bridges, poses a significant challenge. Detection models struggle to simultaneously capture high-resolution details necessary for identifying small targets and low-resolution semantic information essential for recognizing larger objects, leading to unbalanced detection performance across different scales.

To address the aforementioned challenges, this paper proposes a remote sensing image object detection algorithm based on adaptive feature weighting, built upon the Oriented R-CNN framework. The method introduces two core modules: the Adaptive Channel Weighting Module (ACWM) and the Adaptive Multi-Scale Fusion Pyramid (AMFP), aimed at tackling background interference and multi-scale target detection issues. The ACWM module is applied after the STAGE 2 and STAGE 3 phases of the feature extraction network to adaptively adjust channel weights, a critical step for enhancing feature representation. By analyzing statistical correlations between channels using the Correlation-Aware Covariance Weighting (CACW) method, it generates dynamic channel weights to enhance target-related features while suppressing background noise and redundant information, balancing the expression of different channels and thereby improving the feature extraction network's ability to extract features from rotated targets. Subsequently, the AMFP module processes the multi-scale features (C2, C3, C4, and C5) extracted by the feature extraction network, employing a hierarchical approach for feature integration. The Multi-scale Perception and Context Integration Module (MPCIM) fuses global semantic features from deep layers with edge and positional details from shallow layers, effectively addressing the issue of small targets being obscured. The Adaptive Layer Weighting Module (ALWM) optimizes the fusion quality of multi-scale features through adaptive layer-wise weighting, enhancing the model's capability to handle diverse target scales ranging from small vehicles to large ports by dynamically adjusting the contribution of each feature level, thus improving adaptability to various target scales. The optimized features (P2, P3, P4, and P5) are then fed into the Oriented RPN and Oriented RCNN head, which, based on the Oriented R-CNN framework, generates high-precision rotated bounding boxes to complete classification and localization tasks.

To validate the effectiveness of the proposed algorithm, extensive experiments were conducted on the DOTA1.0 and DIOR-R datasets. Ablation studies were performed to assess the contributions of the ACWM and AMFP modules. Results indicate that the ACWM module alone improves the mean Average Precision (mAP) by 0.97% on DOTA1.0 and 1.73% on DIOR-R, while the AMFP module contributes an additional 1.30% and 2.01% improvement, respectively. When both modules are combined, the mAP reaches 78.23% on DOTA1.0 and 68.17% on DIOR-R, representing overall enhancements of 2.36% and 3.87% compared to the baseline Oriented R-CNN model. Visualization comparisons further demonstrate the superiority of the proposed method, showing a significant reduction in missed detections and false positives for small targets, particularly in categories such as Small Vehicle, Swimming Pool, Helicopter, and Storage Tank. The method also exhibits improved localization accuracy and better contour fitting for rotated objects, especially in complex backgrounds with varying angles and dense distributions. The experimental results highlight the robustness and adaptability of the proposed algorithm across diverse remote sensing scenarios.

Compared to state-of-the-art methods such as Faster R-CNN, RoI Transformer, and ARC, the proposed approach achieves superior performance in handling small targets and complex orientations, as evidenced by its higher mAP and enhanced visual detection outcomes. However, limitations remain, including potential performance degradation in extremely cluttered scenes or with ultra-small targets, suggesting avenues for future optimization.

Deep learning  /  Object detection  /  Adaptive feature weighting  /  Optical remote sensing images  /  Multi-scale fusion
李毅, 刘帆, 梁雅婷. 基于自适应特征加权的遥感图像目标检测. 光子学报, 2026 , 55 (2) : 211 -223 . DOI: 10.3788/gzxb20265502.0210003
Yi LI, Fan LIU, Yating LIANG. Remote Sensing Image Object Detection Based on Adaptive Feature Weighting[J]. Acta Photonica Sinica, 2026 , 55 (2) : 211 -223 . DOI: 10.3788/gzxb20265502.0210003
近年来,随着光学遥感技术的快速发展,遥感图像目标检测研究已取得显著进展,并在交通、救援、安全、军事等领域得到广泛应用[1-3]。凭借广阔的视场,遥感图像在大范围监测任务中展现出显著优势[4-6]。然而,遥感图像中的复杂背景信息(如云层、阴影)常会干扰目标特征,削弱检测的鲁棒性。同时,遥感图像中目标尺寸差异较大(如小型车辆与大型港口),使得检测模型难以同时捕捉高分辨率细节和低分辨率语义信息,导致检测精度低、漏检误检率高。
针对背景干扰问题,CHO S J等[7]构建深度去噪自编码器模型,用于消除对抗性背景扰动并恢复干净图像;TIAN Chunwei等[8]通过稀疏模块、特征增强模块、注意力模块和重构模块,提取融合全局与局部特征的显著信息,进而去除复杂背景中的噪声并重建无噪图像;LI Xiaotong等[9]引入特征选择和注意机制,使模型有选择地聚焦于与目标相关的区域,减少无关信息干扰并提升对象的局部感知特征。然而,这些方法大多依赖去噪编码器、注意力机制等外部技术去除特征图背景噪声,却忽略其自身潜力。因特征图部分通道存在冗余或无关信息,目标相关特征的突出度与整体特征表达能力均受限制。
针对多尺度目标检测问题,LIN T Y等[10]设计特征金字塔网络(Feature Pyramid Network, FPN)结构,改变传统目标检测方法依赖顶层特征预测的模式,提出自顶向下的特征图处理方式,并通过横向连接整合浅层和深层网络特征。DONG Xiaohu等[11]将注意力的多级特征融合模块融入FPN,以提升小目标的检测精度。ZHEN Peining等[12]提出在两个FPN中构建多尺度特征融合模块,并结合加权感兴趣区域(Region of Interest, RoI)特征聚合模块,有效融合特征并减少语义信息丢失。GAO Yunxiao等[13]开发基于跨尺度提取的特征超分辨率融合金字塔,显著提高小目标的检测性能。然而,这些方法依赖固定多尺度融合策略,未考虑不同层级特征的重要性差异,无法实现自适应特征加权与融合,导致其在复杂遥感场景中灵活性不足,尤其面对目标尺度剧烈变化时,难以动态优化特征表达。
因此,本文在Oriented R-CNN[14]基础上,提出一种基于自适应特征加权的遥感图像目标检测算法。Oriented R-CNN针对旋转目标检测优化,以ResNet50-FPN为骨干网络,结合定向区域提议网络生成高质量定向候选框,适配遥感图像目标方向多样性;其检测头通过rotated RoIAlign精准提取特征向量,提升定向边界框定位精度,且网络结构可扩展性好,便于嵌入新模块优化性能,无需大规模重构基准框架。本文所提方法通过自适应通道加权模块(Adaptive Channel Weighting Module, ACWM)的自适应通道加权机制,增强目标相关通道,抑制无关信息,并平衡不同通道特征的表达,有效减轻背景噪声对检测的影响;同时,采用自适应多尺度融合金字塔(Adaptive Multi-scale Fusion Pyramid, AMFP),借助多尺度感知与上下文集成模块(Multi-scale Perception and Context Integration Module, MPCIM)以及自适应层间加权模块(Adaptive Layer Weighting Module, ALWM),通过自适应层间加权优化多尺度特征融合质量,动态平衡不同特征图权重,显著提升检测性能。为验证算法的有效性,在DOTA1.0[15]和DIOR-R[16]数据集上开展消融实验和对比实验,全面评估各模块及整体算法的性能提升。
在基线模型Oriented R-CNN的基础上,构建一种基于自适应特征加权的遥感图像目标检测网络,如图1所示,网络由用于提取不同尺度特征的特征提取网络、用于特征融合的自适应多尺度融合金字塔、用于生成高质量候选框的Oriented RPN以及用于候选框分类和精确回归的Oriented RCNN检测头四部分构成。首先,将待检测的遥感图像输入特征提取网络,提取多尺度特征(C2、C3、C4、C5),以满足不同目标尺度的检测需求。在STAGE 2和STAGE 3阶段,网络主要提取目标的边缘、纹理和局部语义特征,但易受复杂背景(如云层、阴影)干扰,为此,应用ACWM模块生成动态通道权重,通过增强目标相关特征,抑制背景噪声,平衡多通道表达,提升网络对旋转目标特征的提取能力。随后,将提取的多尺度特征输入AMFP,结合MPCIM模块和ALWM模块,实现高效特征融合。最后,将优化后的多尺度特征(P2、P3、P4、P5)输入Oriented RPN以生成高质量旋转候选框,再由Oriented RCNN检测头完成分类和精确定位任务。
遥感图像中目标尺度变化大,背景噪声干扰严重,传统通道注意力难以有效抑制冗余特征。为此,ACWM模块通过动态调整通道权重,增强目标相关特征表达,同时抑制背景噪声和冗余信息,如图2所示。该模块应用于特征提取网络的STAGE 2和STAGE 3阶段,这两个阶段主要负责提取目标的边缘、纹理和局部语义特征,但易受复杂背景的干扰。ACWM利用相关性感知协方差加权(Correlation-Aware Covariance Weighting, CACW)方法分析通道间相关性,生成动态通道权重,在增强目标相关特征的同时平衡不同通道的表达能力,从而显著提升特征提取网络对旋转目标特征的提取能力。
为分析通道间的相关性,首先进行特征重塑,将原始特征Fi∈RH×W×C重塑为FiR∈RHW×C,具体是将空间维度展平为样本数,通道维度作为特征数,以便于计算通道间的相关性。接着,使用CACW方法为重塑后的特征FiR生成通道权重αi∈RC×1;最后,将生成的通道权重αi通过逐元素乘法应用于原始特征Fi,优化特征表达。计算过程可以表示为
FiR=Reshape(Fi)
αi=CACW(FiR)
F¯i=Fi⊙αi
式中,F¯i为加权后的特征矩阵,反映了通道相关性调整后的结果,⊙表示逐元素乘法。
传统通道注意力机制通过全局池化生成通道权重,仅关注特征的独立统计特性,却忽略了特征间的动态相关性,尤其在遥感图像中,目标与背景存在复杂交互,亟需更强的依赖性建模。为增强ACWM模块对遥感图像中目标相关特征的提取能力,提出CACW方法,如图3所示。该方法通过协方差矩阵显式建模特征间的全局依赖关系,量化通道间的异质性与冗余性,再通过非线性变换生成自适应权重,用于优化特征选择和调整特征的重要性。
对于输入特征矩阵X∈Rm×n,其中m表示样本数,n表示特征数,CACW的目标是通过自适应调整来生成用于特征选择的权重。首先对X进行去中心化处理,计算中心化特征X¯,随后,计算协方差矩阵Cij,计算过程可以表示为
X˜=X-X¯
Cij=1m-1X˜TX˜
式中,X¯是X中每一列特征的均值,X˜是去中心化后的矩阵,Cij表示特征i和特征j之间的相关性,为后续权重调整提供了基础。
为了消除特征绝对值大小的影响,对Cij进行归一化,得到相关系数矩阵C˜ij。随后,通过多层感知机(Multilayer Perceptron, MLP)对归一化协方差矩阵C˜ij进行非线性变换,生成权重向量γ∈Rn×1,γ用于调整特征的重要性。归一化和非线性变化过程可以表示为
C˜ij=Cij‖X˜i‖‖X˜j‖
γ=f(C˜)
式中,||X˜i||和||X˜j||是特征向量i和j的范数,归一化后的C˜ij表示特征i和特征j之间的标准化相关性,f是一个多层感知机(MLP),来增强特征选择的自适应性。
将特征提取网络提取的多尺度特征输入AMFP,以实现高效的特征融合,如图1所示。首先,借助MPCIM模块,融合深层网络的全局语义特征与浅层网络的边缘、位置特征,既增强模型在复杂背景下检测目标的能力,又抑制浅层网络中的冗余噪声;接着,利用ALWM模块通过自适应层间加权机制,动态调整各层权重以优化多尺度特征的融合质量,进而进一步提升整体检测性能。
MPCIM模块通过融合深层网络的全局语义特征与浅层网络的细节、位置信息,增强对小目标的特征表达,缓解小目标被淹没的问题;同时,该模块抑制复杂背景和噪声干扰,减少边界模糊,进而提升检测精度,如图4所示。
首先对浅层特征图Ci-1和深层特征图Ci+1分别进行降维和上采样操作,以对齐道中间特征图Ci的大小,接着将两个特征图相加进行空间融合得到Pci,计算过程可表示为
Pci=U(Ci+1)⊕D(Ci-1)
式中,U为上采样操作,D为下采样操作,确保空间尺寸一致以实现有效融合。
接着使用大核选择机制,通过双分支大核卷积和空间注意力机制提高模型的动态范围上下文提取能力,使得融合后的特征既能够具有更强的语义特征,又能包含更多的边缘位置信息。最后将其加到中间特征图Ci上,以增强该层的特征表达能力。计算过程可以表示为
b1=C1×1G5,1(Pci)b2=C1×1G7,3G5,1(Pci)
Maski=σi[Cat(b1,b2)]
Pei=Ci⊕C1×1∑i=12(bi⊙Maski)
式中,b1、b2分别表示双分支大核卷积获得的特征,C1×1表示1×1卷积,Gi,j表示卷积核尺寸为i×i步幅为j的卷积,Maski表示空间注意力图,σ表示sigmoid函数,Cat表示特征图拼接,Pei表示最终的融合结果,⊕表示特征拼接,⊙表示逐元素乘法。
ALWM模块通过自适应层间加权机制动态调整各层权重,以此优化多尺度特征的融合质量,如图5所示。首先将浅层特征图Fi下采样到深层特征图的大小,确保二者空间尺寸一致;接着通过空间池化将空间维度H×W压缩为标量FkP,保留通道与层信息以便于分析层间相关性,随后将其拼接得到FP∈RC×N。然后使用CACW方法叠加Softmax操作生成层权重β∈RN×1;最后通过加权求和的方式得到最终的融合特征Fi+1'。计算过程可以表示为
Fi'=D(Fi)
FP=Cat(Pool(Fi+1),Pool(Fi'))
β=Softmax(CACW(FP))
Fi+1'=∑k=1N(Fk⊙β)
式中,D为下采样操作,Pool为空间池化,CACW生成动态权重,Softmax归一化权重,Fk表示Fi+1和Fi'两个不同层的特征。
损失函数采用同基线模型Oriented R-CNN相同的多任务联合损失,总损失函数可以表示为
L=Lcls+Lreg+Lrot
式中,Lcls为交叉熵损失,用于优化目标类别的预测准确性;Lreg为平滑L1损失,用于优化边界框坐标(x,y,w,h)的回归精度;Lrot为旋转角度回归的损失,采用平滑L1损失修正旋转参数,用于优化旋转角度θ的回归精度。
采用公开的DOTA1.0[15]数据集和DIOR-R[16]数据集进行实验,验证所提方法的有效性。
DOTA1.0数据集是遥感图像目标检测领域广泛应用的大型公开数据集,适合测试多尺度特征融合和旋转目标检测的鲁棒性。该数据集包含2806张高分辨率航空图像,分辨率范围从800×800到4 000×4 000像素,涵盖15个常见目标类别,具体包括飞机(PL)、棒球场(BD)、桥梁(BR)、田径场(GTF)、小型车辆(SV)、大型车辆(LV)、船舶(SH)、网球场(TC)、篮球场(BC)、储油罐(ST)、足球场(SBF)、圆形路口(RA)、港口(HA)、游泳池(SP)和直升机(HC)。数据集由多源遥感影像组成,采集自不同传感器和平台,具有多样化的场景、视角和光照条件,共标注了约188 282个目标实例,提供精确的边界框和类别标签。其中,船舶(SH)、小型车辆(SV)和直升机(HC)因目标面积小于32×32像素,被归为小目标。
DIOR-R数据集是专为光学遥感图像目标检测设计的大型公开数据集,适合评估模型的一致性和泛化能力,尤其可验证在光学遥感图像中的小目标适应性。该数据集包含23 463张分辨率为800×800像素的RGB格式遥感图像,涵盖20个常见目标类别,总计192 512个目标实例。具体类别包括飞机(AL)、机场(AT)、棒球场(BF)、篮球场(BC)、桥梁(BG)、烟囱(CM)、大坝(DM)、高速公路服务区(EA)、高速公路收费站(ES)、港口(HB)、高尔夫球场(GC)、田径场(GF)、立交桥(OP)、船舶(SP)、体育场(SD)、储油罐(ST)、网球场(TC)、火车站(TS)、车辆(VM)和风车(WM)。数据集分为训练集(5862张图像)、验证集(5 863张图像)和测试集(11 738张图像)。其中,烟囱(CM)和车辆(VM)等被归为小目标。
采用平均精度(AP)和平均精度均值(mAP)作为模型性能的评价指标。AP是目标检测任务中用于评估模型在单一类别上的性能指标,通过计算Precision-Recall曲线下的面积,综合衡量模型在该类别上的准确率与召回率的平衡关系。具体计算过程为:针对每个类别,在不同置信度阈值下计算准确率和召回率,绘制Precision-Recall曲线,再求取该曲线下的面积,即为该类别的AP值。AP值越高,表明模型在该类别上的检测性能越优。mAP则是评估模型整体性能的核心指标,通过对所有类别AP值取平均计算得出,可综合反映模型在多类别检测任务中的整体表现。
Precision=TPTP+FP
Recall=TPTP+FN
AP=∫01P(R)dR
mAP=1n∑i=1nAPi
式中,TP表示真正例,FP表示假正例,FN表示假负例,P表示精确率,R表示召回率。
为确保实验公平性和可复现性,采用与主流方法一致的数据集处理流程[14,17-19],对比模型均严格遵循其原始论文的配置。所有模型在训练集和验证集上进行训练,并在测试集上开展性能评估。训练流程包括ImageNet[20]预训练和遥感目标检测训练。
在ImageNet预训练阶段,使用MMPretrain[21]工具箱在ImageNet-1K数据集上训练模型,采用AdamW[22]优化器(动量0.9,权重衰减0.05,初始学习率0.001),结合余弦退火和预热策略,共训练300个epoch。在遥感目标检测训练阶段,实验基于MMRotate[23]框架开展,DOTA1.0数据集图像裁剪为1 024×1 024像素补丁(重叠200像素),DIOR-R数据集的输入尺寸为800×800像素;模型针对DOTA1.0训练14个epoch,针对DIOR-R数据集训练16个epoch,训练过程均使用AdamW[22]优化器(权重衰减0.05,初始学习率0.000 2)。测试阶段保持与训练相同的图像分辨率(DOTA1.0为1 024×1 024像素,DIOR-R为800×800像素),且不使用数据增强。所有实验基于PyTorch 2.1框架,在NVIDIA RTX 3090显卡上完成计算。
为验证ACWM模块和AMFP的有效性,在DOTA1.0和DIOR-R数据集上开展消融实验,以评估各模块在特征提取与融合策略中的性能贡献。具体实验设计如下。
1) 基线模型:以Oriented R-CNN作为基准模型,不引入任何额外模块,直接进行遥感图像目标检测。
2) 添加ACWM模块:在Oriented R-CNN的基础上,于特征提取阶段的STAGE 2和STAGE 3阶段后引入ACWM模块,其余部分保持不变。
3) 替换为AMFP:在Oriented R-CNN的基础上,将特征融合网络的FPN替换为AMFP,其他模块保持不变。
4) 联合ACWM和AMFP:在Oriented R-CNN的基础上,同时应用ACWM模块和AMFP,进行遥感图像目标检测。
表1为DOTA1.0数据集上的消融实验结果。其中,基线模型Oriented R-CNN的mAP为75.87%,展示了其在遥感图像目标检测任务中的初始性能水平。在引入ACWM模块后,mAP提升至76.84%,相较基线模型提高了0.97个百分点,表明ACWM模块通过自适应调整特征权重,有效增强了特征提取能力。单独替换特征融合网络FPN为AMFP时,mAP达到77.17%,提升约1.30个百分点,验证了AMFP在多尺度特征融合中的优化作用,能够更好地适应遥感图像中目标尺度的多样性。当同时应用ACWM和AMFP模块时,mAP进一步增至78.23%,相较基线模型总计提升2.36个百分点,可见两模块在特征提取与融合中的协同效应,显著优于单独使用任一模块的效果。
此外,还对所提方法与基线方法进行了可视化对比分析。如图6(a)~(e)所示为基线方法的可视化结果,而图6(f)~(j)则展示了所提模型的可视化效果。通过对比红色虚线圆圈区域的检测效果可明显发现,所提方法显著降低了小目标的漏检和误检现象;同时,该方法在物体角度、位置等信息的定位精度上表现更优,对物体的包裹处理效果更佳,且在部分类别中大幅提升了基线方法的检测性能。
为进一步评估所提出方法对小目标的检测性能,在DIOR-R数据集上开展消融实验。表2为该数据集上的消融实验结果,基线模型Oriented R-CNN的mAP为64.30%,引入ACWM模块后,mAP提升至66.03%,相较基线模型提高了1.73个百分点,表明ACWM在增强目标特征提取方面效果显著;单独将颈部网络FPN替换为AMFP时,mAP达到66.31%,提升约2.01个百分点,验证了AMFP在多尺度特征融合中对小目标检测的优化作用;当同时应用ACWM模块和AMFP时,mAP进一步增至67.58%,相较基线模型总计提升3.28个百分点,且对小目标类别烟囱(CM)和车辆(VM)的mAP提升尤为显著,凸显了两模块在小目标检测中的协同优化作用优于单独使用。这一结果进一步证实了所提策略在复杂遥感场景下的鲁棒性和有效性。
为更直观地对比所提方法与基线模型的性能差异,图7展示了DIOR-R数据集上Airplane类别与Ship类别的PR曲线对比,分别呈现了基线模型及添加模块后模型的曲线结果。通过对比紫色PR曲线(所提方法)与黄色PR曲线(基线方法)可清晰发现,所提方法在精度与召回率两个维度均有显著提升,这一结果进一步验证了其在遥感图像目标检测中的有效性。
在DOTA1.0数据集上,对所提方法与当前先进的检测算法进行了全面比较,结果汇总于表3(其中加粗表示最佳性能,下划线表示次优值)。对比方法包括RetinaNet-O[24]、Faster R-CNN-O[25]、RoI Transformer[26]、DRN[27]、S²ANet[28]、ARS-DETR[29]、DFDet[30]、Oriented R-CNN[14]和ARC[31],既涵盖RetinaNet-O和Faster R-CNN-O等经典方法,也包含ARS-DETR、DFDet等最新SOTA模型。这些模型采用基于锚点、无锚点、旋转检测、多尺度融合等多种技术路线,确保对比的全面性和代表性;且所选模型已在不同数据集上广泛验证,具备较高的引用率与影响力,为评估所提方法的性能优越性提供了可靠基准。此外,模型的提出时间跨度从2017年至2024年,能够反映遥感目标检测领域的最新进展。对比实验涵盖PL(飞机)、BD(棒球场)、BR(桥梁)等类别检测精度及mAP(平均精度)关键指标。实验结果显示,所提方法(Ours)整体mAP达78.23%,相较于RetinaNet-O(68.43%)、Faster R-CNN-O(69.05%)、S²ANet(74.12%)、ARS-DETR(74.16%)、DFDet(74.71%)、Oriented R-CNN(75.87%)RoI Transformer(69.56%)、DRN(70.70%)和ARC(77.35%),表现出显著优势。在具体类别上,所提方法在PL、BD、GTF、SV、LV、SH、TC、SBF、RA、HA和HC类别上,分别取得89.53%、82.92%、77.58%、79.37%、84.87%、88.23%、90.95%、68.20%、70.39%、76.35%和66.68%的最高精度,这一结果得益于ACWM通过相关性感知协方差加权增强目标相关通道,尤其提高了飞机(PL)、船舶(SH)和网球场(TC)等具有显著纹理目标的特征提取能力;同时,AMFP优化了目标的上下文信息,进一步显著提升检测精度。此外,在SH和HC等小目标检测中,所提方法分别达到88.23%和66.68%的精度,展现出卓越的目标适应性。综上,所提方法在检测精度和鲁棒性上均优于现有算法,充分验证了其在复杂遥感场景中的先进性能。
为评估模型的实用性,进一步分析了参数量(Params)与计算复杂度(FLOPs)。结果如表4所示,所提方法相较于对比模型,参数量处于适中水平,计算复杂度较低。这一优势源于ACWM和AMFP模块通过优化特征提取过程,有效减少冗余计算,从而在提升mAP的同时保持了较高的计算效率。
在DIOR-R数据集上,新增ASDNet[32]和DoDet[33]作为对比方法,以进一步验证所提方法的性能优越性和适用性,其中仅RoI Transformer采用ResNet-101作为BackBone,其余方法均采用ResNet-50,表5汇总了所提方法与当前先进检测算法的性能对比结果。结果显示,所提方法(Ours)的整体mAP达67.58%,优于RetinaNet-O(57.55%)、Faster R-CNN-O(59.54%)、ASDNet(61.57%)、DFDet(62.11%)、RoI Transformer(63.87%)、Oriented R-CNN(64.30%)、DODet(65.10%)和ARS-DETR(66.12%)。在具体类别中,所提方法在AL、BC、BG、CM、GC、GF、SD、和VM等类别上,分别取得71.75%、88.04%、45.01%、76.19%、76.58%、83.29%、79.81%和51.86%的最高精度,得益于AMFP的多尺度融合策略,特别对篮球场(BC)和高尔夫球场(GC)等规则目标以及烟囱(CM)等小目标提供了更好的尺度适应性;ACWM的动态通道加权机制有效优化了背景噪声复杂类别的特征表达,进一步提升了模型整体鲁棒性。
在DOTA1.0数据集上,将所提方法与不同检测方法进行可视化检测结果对比,结果如图8所示,其中,基线模型Oriented R-CNN在复杂背景下的检测表现较为局限,尤其在小目标与多角度目标识别中,频繁出现漏检和定位不准的现象。相比之下,所提方法通过自适应通道加权模块(ACWM)和自适应多尺度融合金字塔(AMFP)的协调优化,显著提高了目标的检测率和定位精度,尤其在复杂背景和多角度目标的识别上表现更优。与Faster R-CNN-O、RoI Transformer等其他对比算法相比,所提方法在密集目标分布和角度变化较大的场景中展现出更强的鲁棒性,进一步验证了其在复杂遥感场景中的优越性能。
此外,在DIOR-R数据集上对所提方法进行了可视化分析,结果如图9所示,直观展示了模型在不同场景下的检测效果。可清晰观察到所提方法在小目标检测中显著降低了漏检和误检率,目标定位准确且轮廓包裹效果良好,验证了其在DIOR-R数据集上的适应性。
本文提出了一种基于自适应特征加权的遥感图像目标检测算法。在Oriented R-CNN基础上,设计了自适应通道加权模块,增强目标特征,同时有效抑制背景噪声;引入自适应多尺度融合金字塔,结合多尺度感知与上下文集成模块及自适应层间加权模块,优化多尺度特征融合效果。在DOTA1.0和DIOR-R数据集上的实验结果表明,本文方法的检测精度优于基线模型,消融实验进一步验证了自适应通道加权模块和自适应多尺度融合金字塔在增强鲁棒性与多尺度适应性方面的有效性。尽管研究取得了一定进展,该方法仍存在一定局限性。对于大长宽比的目标易出现漏检问题;在超大分辨率遥感图像处理中,算法的计算复杂度会显著增加,导致其在内存受限设备上的性能下降。在未来的研究工作中,将基于现有研究进一步设计更高质量的定向边界框示方法,尤其针对大长宽比目标,探索更精准的锚框表示方式,以提升其检测精度;同时深入研究如何降低分辨率与数据分布差异对模型推理速度的影响,进一步优化模型的实用性能。

参考文献 引证文献
排序方式:
1
CAI Xinhao, LAI Qiuxia, WANG Yuwei, et al. Poly kernel inception network for remote sensing detection[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024: 27706-27716.
2
JIAO Licheng, WANG Mengjiao, LIU Xu, et al. Multiscale deep learning for detection and recognition: a comprehensive survey[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 36(4): 5900-5920.
3
LI Zongzhe, WANG Qicheng, YANG Weihua, et al. Quantifying the effects of sensor noise and spatial resolution on small target detection in multispectral remote sensing imagery[J]. Acta Photonica Sinica, 2026, 55(1): 0110002.
李宗哲, 王琪程, 杨伟华, 等. 光学遥感图像噪声和空间分辨率对小目标检测精度影响机制研究[J].光子学报, 2026, 55(1): 0110002.
4
GAO Tao, XIA Shilin, LIU Mengkun, et al. MSNet: multi-scale network for object detection in remote sensing images[J]. Pattern Recognition, 2025, 158: 110983.
5
ZHANG Yao, ZHAO Hongyang, LI Xingdong, et al. MERS-Net: a lightweight and efficient remote sensing image object detector[J]. IEEE Geoscience and Remote Sensing Letters, 2025, 22: 1-5.
6
ZENG Xiangjin, LIU Genghuan, CHEN Jianming, et al. A multi-scale hierarchical residual network-based method for tiny object detection in optical remote sensing images[J]. Acta Photonica Sinica, 2024, 53(8): 0810001.
曾祥津, 刘耿焕, 陈建明, 等. 基于多尺度分层残差网络的光学遥感图像微小目标检测[J].光子学报, 2024, 53(8): 0810001.
7
CHO S J, JUN T J, OH B, et al. DAPAS: Denoising autoencoder to prevent adversarial attack in semantic segmentation[C]. 2020 International Joint Conference on Neural Networks (IJCNN), 2020: 1-8.
8
TIAN Chunwei, XU Yong, LI Zuoyong, et al. Attention-guided CNN for image denoising[J]. Neural Networks, 2020, 124: 117-129.
9
LI Xiaotong, JIAO Licheng, ZHU Hao, et al. A collaborative learning tracking network for remote sensing videos[J]. IEEE Transactions on Cybernetics, 2022, 53(3): 1954-1967.
10
LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 2117-2125.
11
DONG Xiaohu, QIN Yao, GAO Yinghui, et al. Attention-based multi-level feature fusion for object detection in remote sensing images[J]. Remote Sensing, 2022, 14(15): 3735.
12
ZHEN Peining, WANG Shuqi, ZHANG Suming, et al. Towards accurate oriented object detection in aerial images with adaptive multi-level feature fusion[J]. ACM Transactions on Multimedia Computing, Communications and Applications, 2023, 19(1): 1-22.
13
GAO Yunxiao, WANG Yongcheng, ZHANG Yuxi, et al. Feature super-resolution fusion with cross-scale distillation for small-object detection in optical remote sensing images[J]. IEEE Geoscience and Remote Sensing Letters, 2024, 21: 1-5.
14
XIE Xingxing, CHENG Gong, WANG Jiabao, et al. Oriented R-CNN for object detection[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021: 3520-3529.
15
XIA Guisong, BAI Xiang, DING Jian, et al. DOTA: a large-scale dataset for object detection in aerial images[C]. Proceedings of the IEEE conference on computer vision and pattern recognition, 2018: 3974-3983.
16
CHENG Gong, WANG Jiabao, LI Ke, et al. Anchor-free oriented proposal generator for object detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-11.
17
LI Yuxuan, HOU Qibin, ZHENG Zhaohui, et al. Large selective kernel network for remote sensing object detection[C]. Proceedings of the IEEE/CVF international conference on computer vision, 2023: 16794-16805.
18
PU Yifan, WANG Yiru, XIA Zhuofan, et al. Adaptive rotated convolution for rotated object detection[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023: 6589-6600.
19
HAN Jiaming, DING Jian, XUE Nan, et al. ReDet: a rotation-equivariant detector for aerial object detection[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 2786-2795.
20
DENG Jia, DONG Wei, SOCHER R, et al. Imagenet: a large-scale hierarchical image database[C]. 2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009: 248-255.
21
MMPRETRAIN CONTRIBUTORS. OpenMMLab's pre-training toolbox and benchmark[EB/OL]. https://github.com/open-mmlab/mmpretrain, 2023.
22
LOSHCHILOV I, HUTTER F. Decoupled weight decay regularization[EB/OL]. arXiv preprint arXiv:2017.
23
ZHOU Yue, YANG Xue, ZHANG Gefan, et al. Mmrotate: a rotated object detection benchmark using pytorch[C]. Proceedings of the 30th ACM International Conference on Multimedia, 2022: 7331-7334.
24
LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]. Proceedings of the IEEE International Conference on Computer Vision, 2017: 2980-2988.
25
REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2016, 39(6): 1137-1149.
26
DING Jian, XUE Nan, LONG Yang, et al. Learning RoI transformer for oriented object detection in aerial images[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 2849-2858.
27
PAN Xingjia, REN Yuqiang, SHENG Kekai, et al. Dynamic refinement network for oriented and densely packed object detection[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 11207-11216.
28
HAN Jiaming, DING Jian, LI Jie, et al. Align deep features for oriented object detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60: 1-11.
29
ZENG Ying, CHEN Yushi, YANG Xue, et al. ARS-DETR: aspect ratio-sensitive detection transformer for aerial oriented object detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 1-15.
30
XIE Xingxing, CHENG Gong, RAO Chaofan, et al. Oriented object detection via contextual dependence mining and penalty-incentive allocation[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 1-10.
31
PU Yifan, WANG Yiru, XIA Zhuofan, et al. Adaptive rotated convolution for rotated object detection[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023: 6589-6600.
32
GONG Maoguo, ZHAO Hongyu, WU Yue, et al. Dual appearance-aware enhancement for oriented object detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 62: 1-14.
33
CHENG Gong, YAO Yanqing, LI Shengyang, et al. Dual-aligned oriented detector[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-11.
2026年第55卷第2期
PDF下载
147
72
引用本文
BibTeX
文章信息
doi: 10.3788/gzxb20265502.0210003
  • 接收时间:2025-08-21
  • 首发时间:2026-10-10
  • 出版时间:2026-02-25
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2025-08-21
  • 录用日期:2025-11-03
基金
作者信息
    1太原理工大学 人工智能学院, 晋中 030600

通讯作者:

参考文献
分享链接
https://castjournals.cast.org.cn/joweb/gzxb/CN/10.3788/gzxb20265502.0210003
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数
科
Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)
属
Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏