中南林业科技大学学报

深度卷积域适应的野生动物图像识别

PDF下载

赵恩庭 ¹ , 张长春 ¹^,²^,³ , 赵海涛 ⁴ , 张军国 ¹^,²^,³

中南林业科技大学学报 | 生物学 2026,46(6): 174-183

收起

中南林业科技大学学报 |生物学 2026 , 46 (6) : 174 -183

深度卷积域适应的野生动物图像识别

全屏

赵恩庭¹, 张长春¹^,²^,³, 赵海涛⁴, 张军国¹^,²^,³

作者信息

^1.北京林业大学　工学院，北京　100083

^2.林木资源高效生产全国重点实验室，北京　100083

^3.北京林业大学生物多样性智慧监测研究中心，北京　100083

^4.陕西省动物研究所，陕西　西安　710032

通讯作者:

张长春（zhangchangchun@bjfu.edu.cn），副教授，博士，硕士研究生导师

张军国（zhangjunguo@bjfu.edu.cn），教授，博士，博士研究生导师。

作者简介:

赵恩庭（zhao_enting@163.com），博士研究生。

Wildlife image recognition with deep convolution domain adaptation

Enting ZHAO¹, Changchun ZHANG¹^,²^,³, Haitao ZHAO⁴, Junguo ZHANG¹^,²^,³

Affiliations

^1.School of Technology, Beijing Forestry University, Beijing 100083, China

^2.State Key Laboratory of Efficient Production of Forest Resources, Beijing 100083, China

^3.Research Center for Biodiversity Intelligent Monitoring, Beijing Forestry University, Beijing 100083, China

^4.Shaanxi Institute of Zoology, Xi'an 710032, Shaanxi, China

出版时间: 2026-06-25 doi: 10.14067/j.cnki.1673-923x.2026.06.017

文章导航

摘要

收起

【目的】

野生动物是生物多样性保护的关键，其基础性工作是野外资源调查，而监测图像的高效识别是野生动物资源科学调查保护的重要前提。然而，由于不同光照、背景、拍摄尺度和物种差异等引发的域偏移问题，易导致识别模型性能下降。本研究旨在提高复杂野外环境下无标签野生动物物种识别的泛化性能，为开放环境野生动物分类研究提供关键技术支撑。

【方法】

本研究提出一种深度卷积域适应的野生动物图像识别模型，以提高模型在无标注场景下的跨域识别准确率。具体地，该模型以预训练的ResNet50网络作为特征提取器，通过最大均值差异约束来捕获野生动物图像的卷积层域不变特征，并设计基于mixup概率分布的特征对齐模块来实现全连接层语义空间映射，提升模型学习高层次语义信息的能力。最后，模型通过熵正则化约束与目标域关联关系挖掘，驱动类别间低密度分离的决策边界优化，进一步提高模型的泛化性和鲁棒性。

【结果】

在分别包含8类野生动物和11类野生动物数据集上进行了一系列实验来验证模型的有效性，实验结果表明，在2个野生动物数据集上获得的平均准确率分别为98.3%和81.7%，与基于对抗学习的基线模型相比，本研究提出的模型对野生动物图像识别性能明显提升。

【结论】

本研究所提出的深度卷积域适应的野生动物图像识别模型能够有效减小因不同时空场景、物种差异引起的域偏移，进而提升跨域野生动物物种识别准确率。

关键词

野生动物 / 图像识别 / 域适应 / 特征对齐

Abstract

收起

【Objective】

Wildlife plays a pivotal role in biodiversity conservation, and field resource surveys form the fundamental basis. Efficient recognition of monitoring images is a critical prerequisite for scientific wildlife resource investigation and protection. However, domain shift issues caused by varying lighting conditions, backgrounds, shooting scales, and species differences often degrade recognition model performance. This study aims to enhance the generalization capability of wildlife species recognition under complex unlabeled field environments, providing key technical support for open-environment wildlife classification research.

【Method】

This paper proposed a deep convolutional domain adaptation model for wildlife image recognition to enhance cross-domain accuracy in unannotated scenarios. Specifically, the model employs a pre-trained ResNet50 network as a feature extractor to capture domain-invariant convolutional features of wildlife images through maximum mean discrepancy (MMD) constraints. A mixup-based probabilistic distribution feature alignment module is designed to map semantic spaces in fully connected layers, enhancing the model's ability to learn high-level semantic information. Finally, entropy regularization constraints and target domain correlation mining are integrated to optimize low-density separation boundaries between categories, further improving the model's generalization and robustness.

【Result】

A series of experiments were performed on two wildlife datasets, which contain 8 and 11 species respectively, to validate the model's effectiveness. The results demonstrate that the proposed model achieves average accuracies of 98.3% and 81.7% on the two datasets, significantly outperforming adversarial learning-based baseline models in wildlife image recognition.

【Conclusion】

The proposed deep convolutional domain adaptation-integrated wildlife image recognition model effectively mitigates domain shifts caused by spatiotemporal scenario variations and species differences, thereby enhancing cross-domain wildlife species recognition accuracy. This approach offers a robust technical framework for advancing wildlife conservation and ecological monitoring.

Key words

wildlife / image recognition / domain adaptation / feature alignment

引用本文

赵恩庭, 张长春, 赵海涛, 张军国. 深度卷积域适应的野生动物图像识别. 中南林业科技大学学报, 2026 , 46 (6) : 174 -183 . DOI: 10.14067/j.cnki.1673-923x.2026.06.017

Enting ZHAO, Changchun ZHANG, Haitao ZHAO, Junguo ZHANG. Wildlife image recognition with deep convolution domain adaptation[J]. Journal of Central South University of Forestry & Technology, 2026 , 46 (6) : 174 -183 . DOI: 10.14067/j.cnki.1673-923x.2026.06.017

基金

收起

国家自然科学基金项目(32371874; 32401569)
北京市自然科学基金项目(6244053)
陕西省创新能力支撑计划(2025JC-GXPT-037)
陕西省科学院基础计划项目(2023K-37)

参考文献

收起

文献

收起

参考文献引证文献

排序方式：

[1]

吴慧, 徐学红, 冯晓娟, 等. 全球视角下的中国生物多样性监测进展与展望[J]. 生物多样性, 2022, 30(10): 196-210.

, XU

X H

, FENG

X J

, et al. Progress and prospect of China biodiversity monitoring from a global perspective[J]. Biodiversity Science, 2022, 30(10): 196-210.

[2]

陈建促, 王越, 朱小飞, 等. 融合多特征图的野生动物视频目标检测方法[J]. 计算机工程与应用, 2020, 56(7): 221-227.

CHEN

J C

, WANG

, ZHU

X F

, et al. Wild animal video object detection method combining multi-feature map[J]. Computer Engineering and Applications, 2020, 56(7): 221-227.

[3]

张军国, 程浙安, 胡春鹤, 等. 野生动物监测光照自适应Retinex图像增强算法[J]. 农业工程学报, 2018, 34(15): 183-189.

ZHANG

J G

, CHENG

Z A

, HU

C H

, et al. Adaptive image enhancement algorithm for wild animal monitoring based on Retinex theory[J]. Transactions of the Chinese Society of Agricultural Engineering, 2018, 34(15): 183-189.

[4]

孙工棋, 张明祥, 雷光春. 黄河流域湿地水鸟多样性保护对策[J]. 生物多样性, 2020, 28(12): 1469-1482.

SUN

G Q

, ZHANG

M X

, LEI

G C

. Wetland water bird biodiversity conservation strategies in the Yellow River basin[J]. Biodiversity Science, 2020, 28(12): 1469-1482.

[5]

ZHANG

C C

, ZHANG

J G

. DJAN: deep joint adaptation network for wildlife image recognition[J]. Animals, 2023, 13: 3333.

[6]

ZHANG

Q Y

, YI

X L

, GUO

J L

, et al. A few-shot rare wildlife image classification method based on style migration data augmentation[J]. Ecological Informatics, 2023, 77: 102237.

[7]

张长春, 李大方, 张军国. 基于Wasserstein距离和相关对齐迁移学习的野生动物图像识别方法[J]. 林业科学, 2024, 60(8): 25-32.

ZHANG

C C

, LI

D F

, ZHANG

J G

. Wildlife images recognition method based on Wasserstein distance and correlation alignment transfer learning[J]. Scientia Silvae Sinicae, 2024, 60(8): 25-32.

[8]

赵恩庭, 张长春, 赵海涛, 等. 基于对抗学习的野生动物图像域适应识别方法[J]. 林业科学, 2025, 61(4): 1-8.

ZHAO

E T

, ZHANG

C C

, ZHAO

H T

, et al. A recognition method of domain adaptation for wildlife images based on adversarial learning[J]. Scientia Silvae Sinicae, 2025, 61(4): 1-8.

[9]

LONG

M S

, CAO

Z J

, et al. Transferable representation learning with deep adaptation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 41(12): 3071-3085.

[10]

SUN

B C

, SAENKO

. Deep coral: correlation alignment for deep domain adaptation[C]//Proceedings of the 14th European Conference on Computer Vision, Amsterdam, Oct 8th-16th, 2016. Amsterdam: Springer, 2016: 443-450.

[11]

TZENG

, HOFFMAN

, ZHANG

, et al. Deep domain confusion: maximizing for domain invariance[J]. arXiv abs, 2014: 1412.3474.

[12]

PENG

X C

, BAI

Q X

, XIA

X D

, et al. Moment matching for multi-source domain adaptation. Proceedings of the 17th IEEE International Conference on Computer Vision, Seoul, Oct 27th-Nov 2nd, 2019. Seoul: IEEE, 2019: 1406-1415.

[13]

TABAK

M A

, NOROUZZADEH

M S

, WOLFSON

D W

, et al. Machine learning to classify animal species in camera trap images: applications in ecology[J]. Methods in Ecology and Evolution, 2019, 10: 585-590.

[14]

YOUSIF

, KAYS

, HE

. Dynamic programming selection of object proposals for sequence-level animal species classification in the wild. IEEE Transactions on Circuits and Systems for Video Technology, 2019, 20.

[15]

B R

, ZENG

Z G

, LIAN

, et al. Few-shot domain adaptation via mixup optimal transport[J]. IEEE Transactions on Image Processing, 2022, 31: 2518-2528.

[16]

GRANDVALET

, BENGIO

, et al. Semi-supervised learning by entropy minimization[C]//Proceedings of the 18th Advances in Neural Information Processing Systems, Vancouver, Dec, 2005. Vancouver: MIT, 2005: 529-536.

[17]

LONG

M S

, CAO

Z J

, WANG

J M

, et al. Conditional adversarial domain adaptation[C]//Proceedings of the 32nd Advances in Neural Information Processing Systems, Montreal, Dec 3rd-8th, 2018. Montreal: Springer, 2018, 31.

[18]

K M

, ZHANG

X Y

, REN

S Q

, et al. Deep residual learning for image recognition[C]//Proceedings of the 29th IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Jun 26th-Jul 1st, 2016. Las Vegas: IEEE, 2016: 770-778.

[19]

RUSSAKOVSKY

, DENG

, SU

, et al. ImageNet large scale visual recognition challenge[J]. International Journal of Computer Vision, 2015, 115: 211-252.

[20]

GANIN

, LEMPITSKY

. Unsupervised domain adaptation by backpropagation[C]//Proceedings of the 32nd International Conference on Machine Learning, Lile, Jul 6th-11st, 2015. Lile: PMLR, 2015: 1180-1189.

[21]

ZHU

Y C

, ZHUANG

F Z

, WANG

J D

, et al. Deep subdomain adaptation network for image classification[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(4): 1713-1722.

[22]

CUI

S H

, WANG

S H

, ZHUO

J B

, et al. Towards discriminability and diversity: batch nuclear-norm maximization under label insufficient situations[C]//Proceedings of the 31st IEEE Conference on Computer Vision and Pattern Recognition, Seattle, Jun 14th-19th, 2020. Seattle: IEEE, 2020: 3940-3949.

[23]

CHEN

, CHEN

H A

, WEI

Z X

, et al. Reusing the task-specific classifier as a discriminator: discriminator-free adversarial domain adaptation[C]//Proceedings of the 33rd IEEE Conference on Computer Vision and Pattern Recognition, New Orleans, Jun 19th-24th, 2022. New Orleans: IEEE, 2022: 7171-7180.

[24]

ZHENG

Z J

, WU

, LYU

L S

, et al. ICCL: independent and correlative correspondence learning for few-shot image classification[J]. Knowledge-Based Systems, 2023, 266: 110412.

2026年第46卷第6期

PDF下载

引用本文

BibTeX

文章信息

doi: 10.14067/j.cnki.1673-923x.2026.06.017

接收时间：2025-06-05
首发时间：2026-10-30
出版时间：2026-06-25

补充材料

相关文章

文章信息

作者

出版历史

收稿日期：2025-06-05
修回日期：2025-08-08

基金

国家自然科学基金项目(32371874; 32401569)

北京市自然科学基金项目(6244053)

陕西省创新能力支撑计划(2025JC-GXPT-037)

陕西省科学院基础计划项目(2023K-37)

作者信息

^1.北京林业大学　工学院，北京　100083

^2.林木资源高效生产全国重点实验室，北京　100083

^3.北京林业大学生物多样性智慧监测研究中心，北京　100083

^4.陕西省动物研究所，陕西　西安　710032

通讯作者:

张长春（zhangchangchun@bjfu.edu.cn），副教授，博士，硕士研究生导师

张军国（zhangjunguo@bjfu.edu.cn），教授，博士，博士研究生导师。

参考文献

分享链接

https://castjournals.cast.org.cn/joweb/znlykj/CN/10.14067/j.cnki.1673-923x.2026.06.017

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

BibTeX
EndNote
RefWorks
TxT