基于融合图像与运动量的奶牛行为识别方法

发布时间：2018-03-25 01:09

本文选题：奶牛行为　切入点：目标分割　出处：《农业机械学报》2017年06期

【摘要】：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,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7 d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、蹄病行为准确率超过80%,发情漏检率最低为3.28%,蹄病漏检率最低为5.32%,提高了规模化养殖管理效率。
[Abstract]:In order to quickly and accurately identify the behaviors that affect the reproduction and health of dairy cows from mass surveillance video, 400 cows in small and lactating cattle houses were used as research objects, and the movement behavior of dairy cows in active areas and on the ramp of milk hall was analyzed. A method of dairy cow object recognition based on image entropy is proposed. By calculating the area of the minimum bounding box and the contour of the target object, the crawling behavior, hoof and back features of the cow are captured in real time, and the moving amount of the identified cow for 7 consecutive days is fused. The experimental results show that the method of this paper is used to monitor the object and movement behavior of dairy cattle in the surveillance video in real time, and to monitor and identify the estrus of dairy cow effectively. The accuracy rate of foot-and-foot disease was more than 80%, the lowest rate of estrus leakage was 3.28%, and the lowest rate of foot-and-foot disease rate was 5.32%, which improved the efficiency of large-scale culture management.
【作者单位】：北京交通大学计算机与信息工程学院;国家农业信息化工程技术研究中心;
【基金】：国家自然科学基金面上项目(61571051)
【分类号】：S823;TP391.41

【相似文献】