Authors

Abstract

The behaviour and posture of animals are closely related to their physiological conditions. To some extent, we can judge their physiological activity by their behaviour and posture. Sheep’s behaviour change significantly during illness or parturition, these behaviours are composed of simple postures, such as standing, eating, lying down. This paper takes the YOLOv3 algorithm as the core technology. It extracts features of sheep’s behaviour and posture through constructing the deep network structure, and uses the pyramid feature fusion and multi-scale prediction to detect the behaviour and posture of sheep. The experimental results show that the training model can effectively detect the three behaviours and postures of sheep: standing, eating, and lying down. The mean average precision is 92.47%. This experiment can be used as a basic technology to judge the physiological activities of sheep. It can be applied to the intelligence of animal husbandry, and has a broad application prospect.

Abstract in Chinese

动物的行为姿态与其生理状况息息相关，提前检测出动物的行为姿态可以在一定程度上判断其生理活动。绵羊在出现疾病或分娩时其行为会有显著的变化，其中这些行为由站立、进食和躺卧简单姿态的组合而成。本文以YOLOv3算法作为核心技术，通过构建深层网络结构，提取绵羊行为姿态特征，采用金字塔特征融合和多尺度的预测对绵羊行为姿态进行检测。经实验表明，在训练集与测试集在9:1的情况下，训练的模型可以有效检测出绵羊的站立、进食、躺卧三种行为姿态。本实验可以作为判断绵羊生理活动的一种基础技术，可以应用到畜牧业智能化中，具有广阔的应用前景。