【摘要】 基于深度图像分析技术研究了一种针对蛋鸡群体行为(分布指数、水平活跃度和垂直活跃度)和群体中个体行为(采食、躺、站和坐)经济简单的自动识别方法。系统由1台3D照相机同步采集数字和深度图像数据,并开发软件进行蛋鸡行为的自动识别,系统5 s采集1次图像数据,共进行10 d的数据采集。描述了行为识别算法并进行了行为识别结果分析。算法对蛋鸡的采食、躺、站和坐的识别准确率分别为90.3%、91.5%、87.5%和56.2%。坐行为识别率较低的原因主要是有时蛋鸡站着探索地面会被误判为坐,这可能与两者之间的分割阈值不够精确有关。更多还原

【Abstract】 Animal behaviors are reflective of its welfare state. They contain important information that can enable producers to better manage livestock. Yet it is more difficult in recognizing the behaviors of group laying hens than other big size animals. Large numbers of hens,homogeneous in appearance,high stocking density and variable body size all contribute to this situation. A computer vision-based system was developed which can automatically recognize group behaviors( distribution index,horizontal activity index and vertical activity index) and individual behaviors( feeding,lying,standing and sitting) of group hens. The system consisted of a 3D camera that simultaneously acquired digital and depth images and a software program that detected and identified the behaviors. The computational algorithm for the analysis of depth images was presented and its performance in recognizing the behaviors as compared with manual recognition was analyzed. The images were acquired at 5 s intervals in 10 d period. The algorithm had the following accuracy of individual behavioral classification: 90. 3% in feeding,91. 5% in lying,87. 5% in standing and 56. 2% in sitting. The lower classification accuracy for the sitting presumably stemmed to imprecise segmentation valve value between sitting and standing and sometimes mistook hen’s standing behavior( exploring in ground) for sitting which could be improved in later test. Hence the reported system provided an effective way to automatically process and classify hen’s group and individual behaviors. This tool was conducive to investigate behavioral responses and time budget of laying hens and facility design and management practice.更多还原