Method for Automatic Identification and Classification of Animal Behavior Based on Deep Learning

  • Jianfeng Li
Keywords: Animal Behavior, Deep Learning, Automatic Recognition, Automatic Classification


Animal behavior is an important basis for people to judge the physiological and psychological
conditions of animals. Therefore, understanding the behavior of animals has important significance for breeding
and managing animals. At present, the main method of analyzing and studying animal behavior is to count and
analyze the results of artificial observation, which is not only high cost but also greatly influenced by subjective
factors. The purpose of this paper is to use the deep learning method to extract the temporal and spatial behavior
characteristics of pigs to realize the automatic identification and classification of pigs’ daily behavior. This paper
firstly preprocesses the video information collected on site, introduces the width parameter based on the classic
SSD model, and designs the sparse depth separable basic network structure. In the prediction module, this paper
proposes a simplified model prediction network structure, and processed the loss function values generated by
the T2 and T3 network layers to achieve the weight update of the model; In the training process, considering the
small sample has certain limitations, designed a training algorithm for autonomously collecting difficult samples.
Finally, defined four kinds of postures of pigs which are standing, lying on the side, lying down and sitting, and
extracted the geometric and Hu-distance posture features, fuses and constitutes a gesture description symbol of a
pig, Constructed SVM posture classifier to realize the pig posture recognition. The research results show that the
algorithm can not only reduce the network parameters, speed up the convergence rate when the pig target
detection model is established, but also fully guarantee the accuracy of data calculation, help to meet the
real-time detection requirements of pigs and improve the automation level of the pig breeding industry.