李道亮,姜国旗,杨建安,白羽,谢琰,王承国.基于计算机视觉的鱼类低氧胁迫行为检测与跟踪算法[J].农业机械学报,2023,54(10):399-406.
LI Daoliang,JIANG Guoqi,YANG Jian'an,BAI Yu,XIE Yan,WANG Chengguo.Detection and Tracking Algorithm of Fish Hypoxia Stress Behavior Based on Computer Vision[J].Transactions of the Chinese Society for Agricultural Machinery,2023,54(10):399-406.
|
摘要点击次数: 193
全文下载次数: 74
|
| 基于计算机视觉的鱼类低氧胁迫行为检测与跟踪算法 [下载全文] |
| Detection and Tracking Algorithm of Fish Hypoxia Stress Behavior Based on Computer Vision [Download Pdf][in English] |
| 投稿时间：2023-03-22 |
| DOI：10.6041/j.issn.1000-1298.2023.10.040 |
| 中文关键词: 加州鲈鱼 胁迫行为 目标检测 多目标跟踪 |
| 基金项目:国家自然科学基金项目（62076244）和山东省科技厅项目（2022LYXZ012） |
|
| 中文摘要:为了能准确检测、跟踪加州鲈鱼因水中溶解氧含量低产生的胁迫行为，本文构建了一种改进的YOLO v5与DeepSORT组合网络算法。在算法方面提出2个改进方案：在原YOLO v5的Backbone和Neck中分别加入2个基于移位窗口的自注意力Swin Transformer模块，提升了网络对目标特征信息的提取能力，以此提升原模型的检测效果；采用Warmup和Cosine Annealing结合的学习率策略，使多目标跟踪算法DeepSORT前期收敛速度更快、更稳定。实验结果表明，在目标检测方面，相对于原YOLO v5，改进的YOLO v5的mAP@0.5、mAP@0.5:0.95和召回率分别提升1.9、1.3、0.8个百分点，在不完全遮挡情况下，改进的算法表现出更好的检测效果。在目标跟踪方面，DeepSORT算法的MOTA、MOTP和IDF1分别提升4.0、0.7、10.7个百分点，并且加州鲈鱼在遮挡前后的ID切换频率得到明显抑制。改进的YOLO v5与DeepSORT跟踪算法更适合于检测、跟踪加州鲈鱼的低氧胁迫行为，能够为加州鲈鱼的养殖提供技术支持。 |
| LI Daoliang JIANG Guoqi YANG Jian'an BAI Yu XIE Yan WANG Chengguo |
| China Agricultural University; Yantai Institute, China Agricultural University |
| Key Words:micropterus salmoides stress behavior target detection multi-target tracking |
| Abstract:In order to accurately detect and track the stress behavior of micropterus salmoides due to low dissolved oxygen content in water, an improved YOLO v5 and DeepSORT combined network algorithm was constructed. In terms of algorithm, two improvement schemes were proposed: two self-attention Swin Transformer modules based on shifted windows were added to the Backbone and Neck of the original YOLO v5, which improved the network's ability to extract target feature information, thereby improving the detection effect of the original model; the learning rate strategy combined with Warmup and Cosine Annealing made the convergence speed of the multi-target tracking algorithm DeepSORT faster and more stable in the early stage. The experimental results showed that in terms of target detection, compared with the original YOLO v5, the mAP@0.5, mAP@0.5:0.95 and recall rate of the improved YOLO v5 were increased by 1.9, 1.3 and 0.8 percentage points, respectively. In the case of incomplete occlusion, the improved algorithm could show better detection results. In terms of target tracking, the MOTA, MOTP, and IDF1 of the DeepSORT algorithm were increased by 4.0, 0.7 and 10.7 percentage points respectively, and the ID switching frequency of micropterus salmoides before and after occlusion was significantly suppressed. The improved YOLO v5 and DeepSORT tracking algorithms were more suitable for detecting and tracking the hypoxic stress behavior of micropterus salmoides, and can provide technical support for the breeding of micropterus salmoides. |
Transactions of the Chinese Society for Agriculture Machinery (CSAM), in charged of China Association for Science and Technology (CAST), sponsored by CSAM and Chinese Academy of Agricultural Mechanization Science(CAAMS), started publication in 1957. It is the earliest interdisciplinary journal in Chinese which combines agricultural and engineering. It always closely grasps the development direction of agriculture engineering disciplines and the published papers represent the highest academic level of agriculture engineering in China. Currently, nearly 8,000 papers have been already published. There are around 3,000 papers contributed to the journal each year, but only around 600 of them will be accepted. Transactions of CSAM focuses on a wide range of agricultural machinery, irrigation, electronics, robotics, agro-products engineering, biological energy, agricultural structures and environment and more. Subjects in Transactions of the CSAM have been embodied by many internationally well-known index systems, such as: EI Compendex, CA, CSA, etc.
|
| 下载PDF阅读器 |
|
|
|
