Abstract: This study proposes a target detection and tracking method based on the improved YOLOv11 model—YOLOv11n-DFM. It aims to evaluate losses during the fishing process by detecting the number of crab traps being lifted or lowered and to assess the normalcy of the trap mechanism by detecting the number of traps in key areas. The method integrates DyHead, FocalModulation, and CCFM modules into the YOLOv11n model to enhance multi-scale feature fusion, improve detection accuracy for traps of different scales, and reduce computational and memory costs. Additionally, the ByteTrack algorithm is employed to ensure precise tracking of the traps. Experimental results demonstrate that the YOLOv11n-DFM model improves detection accuracy by 1%, increases mAP@50-95 by 0.8%, while mAP@50 and recall remain unchanged. Compared to the YOLOv11n model, the detection performance is enhanced while maintaining the same detection efficiency. The study indicates that the YOLOv11n-DFM model excels in detecting and tracking the crabs' traps, consumes fewer computational resources, and is suitable for deployment in environments with limited computing power. It provides valuable references for fishery monitoring, resource management, and the future automation of crab trap deployment and collection.

Key words: crab trap, YOLOv11, detection and tracking, bytetrack

摘要： 为检测梭子蟹蟹笼的上笼下笼数量来评估捕捞过程中的损失，提出了一种基于改进YOLOvn11模型的目标检测与追踪方法YOLOv11n-DFM，通过对关键区域蟹笼数量的检测判断蟹笼上下笼装置是否正常。该方法在YOLOv11n模型的基础上集成了DyHead、FocalModulation和CCFM模块，以增强模型的多尺度特征融合能力，提升对不同尺度蟹笼的检测精度，并减少计算量和内存消耗。同时，采用ByteTrack算法实现对蟹笼的精确追踪。试验结果表明，YOLOv11n-DFM模型在检测准确率上提高了1%，mAP@50-95提升了0.8%，而mAP@50和召回率保持不变，且相较于YOLOv11n模型，检测性能得以提升的同时，保持了相同的检测效率。研究表明，YOLOv11n-DFM模型在梭子蟹蟹笼的检测与追踪任务中表现出色，消耗较少计算资源，适合在算力有限的环境中部署，为渔业监测、资源管理和未来自动化蟹笼排放与收集提供了参考。

关键词: 蟹笼, YOLOv11, 检测追踪, ByteTrack