基于YOLO 11n-LD和Strongsort-LBM的多目标奶牛跟踪方法

Multi-target Dairy Cow Tracking Method Based on YOLO 11n-LD and Strongsort-LBM

摘要:

随着现代化畜牧养殖发展，奶牛行为监测对于降低养殖成本、疾病预防等具有重要意义，而多目标跟踪技术是其实现的关键。针对奶牛行为分析中多目标跟踪技术需求，本研究提出了一种基于YOLO 11n-LD和Strongsort-LBM的奶牛多目标跟踪方法，旨在解决现有视觉算法存在ID切换的问题。在目标检测方面，引入LDConv、SimSPPF和SIoU损失函数，提升检测精度，构建了目标检测模型YOLO 11n-LD，其精确率、召回率和mAP@0.5分别达到96.1%、95.1%和95.7%。在跟踪优化方面，改进Strongsort算法，设计了基于位置的匹配机制，结合改进的匈牙利匹配算法对失配轨迹与检测目标进行再关联。经测试，Strongsort-LBM算法HOTA、MOTA、MOTP和IDF1分别为83.43%、96.92%、79.95%和89.90%，ID切换次数降至18次，较YOLO 11n+Strongsort算法低约86%，且在HOTA、IDF1上优于Deepsort、ByteTrack和Ocsort算法，显著减少了ID切换次数。研究结果为牛舍养殖环境下奶牛行为监测提供了高效可靠的跟踪解决方案。

Abstract:

With the development of modern animal husbandry, dairy cow behavior monitoring is of great significance for reducing breeding costs and preventing diseases, and multi-object tracking technology is the key to its realization. Aiming at the demand for multi-object tracking technology in dairy cow behavior analysis, a dairy cow multi-object tracking method was proposed based on YOLO 11n-LD and Strongsort-LBM to solve the ID switching problem of existing visual algorithms. In the aspect of object detection, LDConv, SimSPPF and SIoU loss functions were introduced to improve detection accuracy, and the object detection model YOLO 11n-LD was constructed, whose precision, recall and mAP@0.5 reached 96.1%, 95.1% and 95.7%, respectively. In terms of tracking optimization, the Strongsort algorithm was improved, a position-based matching mechanism was designed, and the mismatched trajectories and detection targets were re-associated combined with the optimized Hungarian matching algorithm. The test results showed that the HOTA, MOTA, MOTP and IDF1 of Strongsort-LBM algorithm were 83.43%, 96.92%, 79.95% and 89.90%, respectively, and the number of ID switches was reduced to 18, which was about 86% lower than that of YOLO 11n+Strongsort algorithm. It was superior to DeepSORT, ByteTrack and Ocsort algorithms in HOTA and IDF1, and greatly reduced the number of ID switches. The research result can provide an efficient and reliable tracking solution for dairy cow behavior monitoring in cattle shed breeding environment.