With the rapid development of the camel milk industry and tourism, camel farming is gradually emerging. The basic behaviors of camels (such as standing, walking, grazing, and resting) are important indicators of their health status and welfare level. Timely monitoring and analysis of these behaviors are crucial for assessing the physiological state of camels and optimizing their management. To achieve automated recognition of camel behaviors, this paper constructed a dataset of basic camel behaviors under free-grazing conditions and designed DAF-Net, a network built upon the Yolov12 framework. In the feature extraction stage, DAF-Net employs Dfe-Net (Dynamic feature extraction Net) for efficient representation learning. Three dynamically adaptive modules (C3BA, A2Dy, and AGU) were integrated to further enhance overall performance. In addition, during the data processing stage, The G-Edge SSIM algorithm is proposed in this paper to address the issue of excessive similarity between consecutive frames, and it operates without the need for GPU computational resources. Experimental results demonstrate that the proposed method achieved excellent recognition accuracy for four basic behaviors—resting, grazing, standing, and walking—at 99.0%, 96.8%, 89.5%, and 89.2%, respectively, with an overall accuracy of 93.6%. Moreover, the method enables multi-camel behavior recognition in video sequences, providing a feasible approach for camel health assessment and welfare monitoring. This study offers new insights into intelligent camel farming management and biodiversity development in free-grazing pastures of arid regions.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号