Scholarlink

<p>The Chinese mitten crab is one of China’s main aquaculture species. Tracking and calculating activity levels can reflect its health status and behavioral characteristics, aiding in improving farming efficiency. However, due to the high similarity in dorsal shell appearance, tracking algorithms face issues such as low accuracy and frequent ID switches. This study proposes a Chinese mitten crab tracking method based on YOLOv11-BoT-SORT, including target detection, target tracking, and activity level estimation. In the detection stage, YOLOv11 serves as the baseline network, with the MobileOne convolution module introduced to reduce computation and parameters; the EMA (Excitation and Modulation Attention) mechanism is added to enhance dorsal shell feature extraction, maintaining model accuracy. In the tracking stage, the BoT-SORT algorithm is used, with ResNeST50 network extracting dorsal shell appearance features to improve tracking precision and reduce ID switches. For activity level estimation, the crab’s speed, turning angle, moving distance, and activity time ratio within a specified period are calculated, then combined to determine activity level. Experiments on a self-built dataset show a detection mAP50 of 99.1%, model size of 7.03&#xa0;MB, a 28.8% reduction compared to the original YOLOv11. The tracking method achieves a MOTA of 88.138%, with IDSW reduced to 3. The average activity level of the Chinese mitten crab was 0.45, with a standard deviation of 0.18, exhibiting an approximately normal distribution. This research enables detection, tracking, and activity level estimation of Chinese mitten crabs, which could have potential applications for aquaculture producers.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Crab tracking and activity estimation method based on YOLOv11-BoT-SORT

  • 期刊论文

  • 发表日期：

    2026年7月18日

查看全文
  • Fu Jiayi,
  • Kong Mingrui,
  • Pan Hongyu,
  • Li Chenlong,
  • Duan Qingling

摘要

The Chinese mitten crab is one of China’s main aquaculture species. Tracking and calculating activity levels can reflect its health status and behavioral characteristics, aiding in improving farming efficiency. However, due to the high similarity in dorsal shell appearance, tracking algorithms face issues such as low accuracy and frequent ID switches. This study proposes a Chinese mitten crab tracking method based on YOLOv11-BoT-SORT, including target detection, target tracking, and activity level estimation. In the detection stage, YOLOv11 serves as the baseline network, with the MobileOne convolution module introduced to reduce computation and parameters; the EMA (Excitation and Modulation Attention) mechanism is added to enhance dorsal shell feature extraction, maintaining model accuracy. In the tracking stage, the BoT-SORT algorithm is used, with ResNeST50 network extracting dorsal shell appearance features to improve tracking precision and reduce ID switches. For activity level estimation, the crab’s speed, turning angle, moving distance, and activity time ratio within a specified period are calculated, then combined to determine activity level. Experiments on a self-built dataset show a detection mAP50 of 99.1%, model size of 7.03 MB, a 28.8% reduction compared to the original YOLOv11. The tracking method achieves a MOTA of 88.138%, with IDSW reduced to 3. The average activity level of the Chinese mitten crab was 0.45, with a standard deviation of 0.18, exhibiting an approximately normal distribution. This research enables detection, tracking, and activity level estimation of Chinese mitten crabs, which could have potential applications for aquaculture producers.