Scholarlink

<p>In intensive high-density recirculating aquaculture systems, real-time detection of shrimp vitality and death time estimation are critical technical challenges for ensuring production efficiency and product quality. The decay of dead shrimp rapidly pollutes water, making timely death time determination essential to pinpoint pollution sources, reduce the time dead shrimp remain in the water, and mitigate the risk of mass mortality. This study proposes a multi-task monitoring method based on an improved YOLOv11 framework, aiming to enable non-destructive, precise discrimination of shrimp vitality and dynamic prediction of time of death. The core contributions of this research include the following: First, an image dataset comprising three shrimp vitality—living, half-dead, and dead—was constructed, and an Enhanced Feature Correlation Fusion (EFC) module was introduced to optimize the multi-scale feature fusion capacity of the YOLOv11 model. This enhancement enables high-accuracy object detection under complex underwater conditions, achieving an mAP@0.5 of 0.941 and an F1-score of 0.937 on the test set. Second, this study incorporates keypoint detection techniques to accurately capture the spatial dynamics of the shrimp’s head, mid-body point, and tail endpoint, thereby calculating the curling angle that characterizes the progression of rigor mortis. To investigate the influence of environmental factors, comparative experiments were conducted in parallel under saline and freshwater conditions. Results indicate that salinity differences significantly affect curling rates through osmotic mechanisms, with freshwater environments slowing the process. Finally, a time-of-death prediction model constructed via polynomial regression achieved a mean absolute error of 6.3&#xa0;min between predicted values and actual recorded times. By integrating computer vision with biomechanical principles, this study provides an intelligent monitoring solution that unifies state recognition and temporal inference for industrialized aquaculture systems.</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

Research on a multi-task monitoring method for shrimp vitality assessment

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Yufeng Xie,
  • Lei Wang,
  • Xin Wu,
  • Shuangle Wu,
  • Weiming Cai

摘要

In intensive high-density recirculating aquaculture systems, real-time detection of shrimp vitality and death time estimation are critical technical challenges for ensuring production efficiency and product quality. The decay of dead shrimp rapidly pollutes water, making timely death time determination essential to pinpoint pollution sources, reduce the time dead shrimp remain in the water, and mitigate the risk of mass mortality. This study proposes a multi-task monitoring method based on an improved YOLOv11 framework, aiming to enable non-destructive, precise discrimination of shrimp vitality and dynamic prediction of time of death. The core contributions of this research include the following: First, an image dataset comprising three shrimp vitality—living, half-dead, and dead—was constructed, and an Enhanced Feature Correlation Fusion (EFC) module was introduced to optimize the multi-scale feature fusion capacity of the YOLOv11 model. This enhancement enables high-accuracy object detection under complex underwater conditions, achieving an mAP@0.5 of 0.941 and an F1-score of 0.937 on the test set. Second, this study incorporates keypoint detection techniques to accurately capture the spatial dynamics of the shrimp’s head, mid-body point, and tail endpoint, thereby calculating the curling angle that characterizes the progression of rigor mortis. To investigate the influence of environmental factors, comparative experiments were conducted in parallel under saline and freshwater conditions. Results indicate that salinity differences significantly affect curling rates through osmotic mechanisms, with freshwater environments slowing the process. Finally, a time-of-death prediction model constructed via polynomial regression achieved a mean absolute error of 6.3 min between predicted values and actual recorded times. By integrating computer vision with biomechanical principles, this study provides an intelligent monitoring solution that unifies state recognition and temporal inference for industrialized aquaculture systems.