【摘要】 针对猪易发生粘连而难以分割的问题，利用改进的YOLOV4网络用于猪个体检测，将双边滤波改为三通道的RGB-双边滤波，提升边缘锐化效果，在二维Otsu阈值分割算法中引入类内与类间融合的分割度量函数，提升分割效果。选取的有锐化边缘和没有锐化边缘图片做对比实验，实验表明，添加锐化边缘后粘连猪的分割更清晰。选取单个猪、多只猪不粘连和多只猪粘连三种数据集做对比实验，实验表明，添加锐化边缘后的分割平均准确率都较高，并且最大与最小准确率的跨度不大，表明该方法的稳定性较好。更多还原

【Abstract】 Aiming at the problem that pigs are prone to adhesion and difficult to segment, the improved YOLOV4 network is used for pig individual detection. The bilateral filter is changed to a three-channel RGB-bilateral filter to improve the edge sharpening effect. The segmentation metric function of intra-class and inter-class fusion is introduced in the two-dimensional Otsu threshold segmentation algorithm to improve the segmentation effect. The images with and without sharpened edges are selected for comparative experiments, and the results show that the segmentation of the sticky pigs is clearer after adding the sharpening edges.The comparative experiments of the three datasets of single pig, multiple pigs non-adhesion and multi-pig adhesion show that the average accuracy of the segmentation after adding sharpened edges is high, and the span between the maximum and minimum accuracy rates is not large. It shows that the stability of the method is better.更多还原