机器视觉识别单只蛋鸡行为的方法

动物行为是一个重要的动物福利评价指标。为了实现对蛋鸡行为的自动监控,该文提出了利用计算机视觉技术对单幅蛋鸡图像进行行为识别的方法,可自动识别单只蛋鸡的运动、饮水、采食、修饰、抖动、休息、拍翅膀、探索、举翅膀的行为,并可长时间追踪蛋鸡的活动情况及运动轨迹。运动、采食和饮水通过追踪蛋鸡的位移和位置直接识别;拍翅膀、修饰、休息、探索、抖动、举翅膀则使用贝叶斯分类法基于10个特征量进行识别,所引入的蛋鸡上下轮廓到最小二乘拟合椭圆长轴距离的相关系数可有效追踪蛋鸡头部,从而提高了修饰和探索的识别率。对9219幅图像进行蛋鸡行为识别的识别率分别为:运动99.4%、饮水80.7%、采食87.3%、修饰81.6%、抖动69.8%、休息86.2%、拍翅膀100%、探索(包括啄食)54.0%、举翅膀64.6%。

Behavior recognition method for individual laying hen based on computer vision

Animal behavior is an important animal welfare evaluation index. In order to achieve the automatic monitoring of laying hens behavior, a method of an individual laying hen behavior recognition based on computer vision technology was proposed in this study. The behaviors of moving, drinking, feeding, preening, shaking, resting, wing flapping, exploration and wing lifting of individual laying hen in single image could be distinguished by tracing the displacement and location of the laying hen, and the track activities and movement trajectories in a long time could be tracked using Bayesian classification with ten parameters calculated from the hen images using image processing techniques. The correlation coefficient of distance from laying hen images upper and lower contour to least-squares fitting ellipse longitudinal could effectively track the hens head, and the recognition rate of grooming behavior and exploratory behavior were improved. The test results of 9219 images indicated that the recognition rates of moving, drinking, feeding, preening, shaking, resting, wing flapping, exploration and wing lifting were 99.4%, 80.7%, 87.3%, 81.6%, 69.8%, 86.2%, 100%, 54.0% and 64.6%, respectively.