Full
Outline
As the body surface mapping area of human organs and meridians, the accurate identification and location of auricular points in traditional Chinese medicine is the key technical support for auricular acupuncture diagnosis and treatment, traditional Chinese medicine health monitoring and other scenes. Traditional ear acupoint detection is highly dependent on the clinical experience of professional doctors, which is not only subjective and inefficient, but also difficult to meet the needs of standardized application. The existing target detection algorithms generally have pain points that are difficult to balance accuracy and speed in auricular point detection. To this end, this study proposes an improved traditional Chinese medicine auricular point detection algorithm based on YOLOv12. First of all, combined with the national standard of auricular points and 9 kinds of TCM diseases, a special data set containing 31 auricular points partitions was constructed, including 2140 auricular points images covering different age groups, ear shape characteristics and shooting environment. In terms of model optimization, the C3k2-GV3S Lightweight Residual Feature Aggregation Module is introduced to replace the C3k2 module of the original backbone network. With the help of the two-stage strategy of “basic feature generation + cheap phantom feature completion”, the parameter quantity and calculation amount are reduced by about 50% on the premise of retaining the advantage of residual feature multiplexing, and the key feature representation of small auricular targets is strengthened. The bilinear adaptive boundary-enhanced upsampling module BABEU (Bilinear Adaptive Boundary-Enhanced Upsampling Module) is designed to replace the traditional interpolation operation. Through the combination strategy of “bilinear interpolation smoothing feature + learnable convolution thinning boundary”, the high-frequency detail features of the auricular point boundary are adaptively learned to avoid pixel block effect and checkerboard artifacts, and the boundary feature recovery accuracy of the auricular point small target is significantly improved. Finally, a bilinear adaptive boundary enhanced upsampling and lightweight residual feature aggregation network YOLOv12n-BiGS (Bilinear Adaptive Boundary-Enhanced Upsampling and Lightweight Residual Feature Aggregation Network) based on YOLOv12n has been constructed. The experimental results show that the average Dice coefficient of the model on the dedicated data set is 0.5821, and the reasoning time is only 3.3 ms. The comprehensive performance is better than the mainstream lightweight algorithms such as YOLOv8 n, YOLOv11 n and DeepEar. The special data set constructed in this study meets the clinical needs, and the improved model realizes the collaborative optimization of auricular point detection accuracy and real-time performance, which provides technical support for the intelligent and standardized promotion of auricular point diagnosis and treatment in traditional Chinese medicine.
74
12
| 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
No. 86 Xueyuan South Road, Haidian District, Beijing
010-62662699(Operation and Technical Support)
qkjq@cast.org.cn
Copyright © 2025 China Association for Science and Technology. All rights reserved. For all open access content, the relevant licensing terms apply.
Sponsored by the Office of the Leading Group for Cybersecurity and Informatization of CAST, and supported by Science and Technology Review Publishing House
