Key points are not available for this paper at this time.
Abstract Daily behavior detection and monitoring of sheep is crucial for assessing their health status. In recent years, computer vision has been widely used in livestock behavior detection, but it usually requires large memory and computational resources. In addition, most studies have focused only on the behavior of sheep during the day, while the behavior of sheep during the night is equally important for a comprehensive understanding of their health status and well-being. Therefore, in this study, we developed a lightweight daily behavior detection and counting method for housed sheep to detect lying, feeding, and standing behaviors, and to count the number of each behavior as well as the total number of sheep. First, we propose a new PCBAM module and incorporate it into the neck part of YOLOv8n to enhance the feature information contained in the feature map, second, we use the slim neck design paradigm incorporating GSConv to lighten and improve the model operation efficiency, and finally, we reconstruct the detection head to eliminate the redundant small target detection head, reduce the model computational burden, and improve the detection performance of medium and large targets. The Sheep-YOLO model is validated using the daily behavioral dataset of housed sheep, and the experimental results show that the improved model is effective in detecting sheep behavior in complex environments, and the mAP@0.5 is improved by 5.4% compared to the baseline model, and in particular, the lying and feeding behaviors of sheep are improved by 7.2% and 8.8%, respectively. Comparative experiments with other mainstream target detection algorithms validate the advantages of our proposed model for sheep behavior detection. This study provides an effective solution for behavioral detection and counting of housed sheep.
Wang et al. (Tue,) studied this question.