Recognition Of Individual Pigs Based On Gabor Directional Histogram And Support Vector Machine

In recent years,machine vision-based identification systems have been widely used in various scenes,and modern large-scale pig farms have made new demands for real-time monitoring of pig growth activities and timely detection of abnormal behavior.Recognition of individual pig is a prerequisite for the pig's behavior study.In order to make better use of image processing technology to monitor the growth activities of piglets,this paper proposes a method of pig individual identification based on Gabor direction histogram and Characteristics of pig hair pattern.For the non-rigid body characteristics of pig,the external shape of the posture will change with the characteristics of change.In order to obtain a stable feature region,the key points of the pig body were found on the outline of the pig by using the key detection algorithm of the pig.Then,two characteristic regions of shoulder and buttocks were divided on the contours of pigs based on these key points.The two characteristic areas were less affected by pig deformation and remained relatively stable.In this paper,the characteristics of hair patterns in the body of the pig are rich in direction and density.In order to make full use of these characteristic information,we use the Gabor filter to perform multi-directional,multi-directional filtering on the feature region after the feature region is determined.In order to make full use of the position information of hair features,this paper uses local descriptors rather than global descriptors to describe the hair features,because there are very different hair patterns in different parts of pigs.The specific approach is to further mesh the feature area,and then use the zero-DC response of the filter output in each grid to divide the Gabor direction histogram.In this paper,the Gabor direction histogram obtained from each pig in the sample image is extracted from the sample database.After the establishment of the sample library,we use the SVM based on the combined kernel function to train the sample features in the sample library to generate the classifier.Finally,the Gabor direction histogram of the pigs in the test sample image is input to the classifier for identification.In order to test whether the hair pattern feature proposed in this paper and Gabor direction histogram feature extraction method are feasible and effective for pig identification,the cumulative matching characteristic curve(CMC)is used as the performance evaluation index.The algorithm proposed in this paper and the traditional texture feature extraction algorithm LBP,HOG were compared experimentally.The experimental results show that the algorithm has the highest recognition rate of 86.51%.At the same time,the algorithm has the lowest feature dimension.In the last part of the experiment,we tested the algorithm using different resolution pictures.In the picture resolution from 1760 × 1840 down to 880 × 920 when the algorithm can maintain 83.21% recognition rate,only when the resolution down to 440 × 460 when the recognition rate dropped significantly.The experimental results show that the algorithm can achieve good recognition effect under the condition of low resolution picture,and has good practicability.This study provides a new idea for the individual identification of pigs without stress,and also provides some technical support for further exploration of individual behavior analysis.