【摘要】 计算机视觉技术越来越多地应用于生猪饮水等行为识别中,以判断生猪健康状况。现有的饮水识别方法主要依赖目标轮廓,而传统的阈值分割方式受光照、噪点等因素影响较大,提取的轮廓不够精确。提出一种基于深度学习目标检测算法YOLO(You Only Look Once,YOLO)的生猪行为识别方法,根据生猪位置与饮水区的关系以及是否处于静止状态综合判断其饮水行为。该方法不依赖目标轮廓,且无复杂的手动特征提取过程。在深度学习框架tensorflow上进行群养猪检测、定位以及饮水行为识别。实验证明,该算法比基于轮廓的饮水识别算法精度提高3%,达到94.0%。更多还原

【Abstract】 Computer vision technology has been increasingly used in pig drinking water behavior recognition to judge the health status of pigs. The existing drinking water recognition methods mainly rely on the target contour,while the traditional threshold segmentation method is greatly affected by light,noise and other factors,and the extracted contour is not accurate enough. Therefore,this paper proposes a pig detection algorithm based on the deep learning target detection algorithm Yolo(you only look once),which identifies the location of the pig,and then comprehensively judges the drinking behavior according to the relationship between the pig location and the drinking water area as well as whether it is still or not. This method does not depend on the contour of the target,and does not need complex manual feature extraction. In this paper,the in-depth learning framework tensorflow is used to detect and locate pigs and identify drinking behavior. Experiments show that the accuracy of this algorithm is 3% higher than that of the contour based algorithm,and it reaches 94.0%.更多还原