论文部分内容阅读
为从海量监控视频中快速、准确识别影响奶牛繁殖与健康的行为,本文以小育成牛舍与泌乳牛舍中400头奶牛为研究对象,分析了奶牛在活动区与奶厅匝道的运动行为,提出了一种基于图像熵的奶牛目标对象识别方法,通过最小包围盒面积计算与目标对象轮廓图,实时捕获奶牛爬跨行为与蹄部、背部特征,融合被识别奶牛连续7d的运动量,判断影响奶牛健康繁殖的异常行为。试验结果表明,利用本文方法对监控视频内奶牛目标对象、运动行为进行实时监测,有效监控识别奶牛发情、体病行为准确率超过80%,发情漏检率最低达到3.28%,蹄病漏检率最低为5.32%,提高了规模化养殖管理效率。
In order to quickly and accurately identify the effects of reproductive and health effects on dairy cattle from surveillance video, 400 dairy cows in small cowshed and lactating cowshed were selected to study the movement of dairy cows in the activity area and the ramp of dairy cows. A method of target recognition based on image entropy was proposed. Through the calculation of the minimum bounding box area and the contour map of the target object, the crawling behavior and the characteristics of the hoof and back of the cow were captured in real time. Abnormal behavior of healthy reproduction of dairy cows. The test results show that using this method to monitor the target and movement of cows in the video in real time, effectively monitoring and controlling estrus, the correct rate of body disease is more than 80% and the rate of estrus missing is 3.28% A minimum of 5.32%, improve the efficiency of large-scale farming management.