Pig Posture Classification And Abnormal Behavior Analysis Based On Decision Tree Support Vector Machine

With the improvement of the degree of automation and intelligence in pig farmingindustry, the machine vision technology and intelligent monitoring technology have been thenew direction of application. More and more researchers have focused on this hot spot andstarted with some research. In the previous period of pig farming industry, the abnormality ofthe pig was usually detected by breeder’s observation, it cost plenty of time and manpower,and the result was easily influenced by subjective experience. Considering this, usingmachine vision technology and intelligent monitoring technology to replace the manual workwill be the trend of pig farming industry in the future. In this background, this paperpresented a classification algorithm of pig postures and an analysis system of abnormality ofpig, based on geometric parameter features and Decision Tree Support Vector Machine(DT-SVM).After introduce the overseas and domestic research status of classification of pig postureand evaluation of abnormality of pig, this paper is presented as follows, acquisition,processing, classification, recognition, and analysis of abnormality. The main content of thispaper is as follows.Firstly, considering the fact that the pig moves slowly, this paper uses the combination ofCCD industrial camera and video server to capture the RGB image of pig postures. Then,uses the object extraction algorithm by combining color and texture information,morphological image processing and the median filter algorithm to get the whole, smoothbinary image of the pig. Secondly, this paper analyzes the types of pig postures, chooses five postures, includinglying, side view standing with raised head, side view standing with lowered head, side viewstanding horizontally and front view standing. Then analyzes the image features which coulddescribe the pig posture effectively, finally chooses the geometric parameter feature of shapefeature among the color feature, texture feature, shape feature and spatial relationship feature.The specific features are as follows, circularity, aspect ratio, elongation, ratio of the height ofcentroid and height, ratio of the height of centroid and left distance, left-right ratio of centroid,left-right ratio of peak point, left (right) angle cosine and left (right) raised head degree,11features totally which could describe the postures.Thirdly, after the acquisition of11dimensions of features, the paper uses the SupportVector Machine (SVM) to classify the features. This paper applies the Decision TreeMulti-Classification Support Vector Machine (DT-SVM) to the classification, designed thedecision tree for the pig posture classification. Amount of experiences shows that the11dimensions of features could describe the pig postures effectively, the mean accuracy is over90%.Finally, based on the pig posture classification model which was built before, this paperachieves the pig posture image annotation, which could describe the image sequence of pigpostures. The annotation could build a connection between the time and posture. Then, importthe Behavior Anchored Rating Scale (BARS), which was a method of management, to buildan evaluation system based on posture and time. The evaluation system has referencesignificance.This method of the classification of pig posture and evaluation of abnormality of the piggives a new idea to the development of pig farming industry. It could help to reduce themanual labor, to discover the abnormality of pig and to avoid the loss.