A Feature Region Identification Method Of Pigs Based On Chordal Axis Transform

As an important methods to improve the level of informatization in traditional breeding industry,computer vision technology has been increasingly applied to large-scale breeding of pigs.When analyzing and monitoring individual behaviors of pigs,we must first identify the corresponding feature regions of pigs,and then analyze the posture and behavior of pigs.Therefore,accurate identification of feature regions is crucial.In this paper,a method based on chordal axis transform is used to identify the feature regions(head,ear,nose,leg,and tail)of the pig under top view conditions.This will provide a new approach for follow-up research on abnormal behavior analysis of pigs.After using two threshold segmentation methods to obtain contours of individual pigs,a corner point detection algorithm based on the curvature scale space is used to extract key points on the boundary of the pig,and these points are at key positions,such as the ear root,tail root etc.And discrete points are selected evenly between these key points,and then the method based on Delaunay is used to triangulate the pig body.A series of triangles could be obtained,and the strength values of the common edges(chords)of the two connected triangles are calculated.If the strength value is small,the probability that the two triangles belong to the same region is large,thus they can be combined.With this characteristic of the chord strength value,triangles belonging to the same region could be merged.Thus,the pig is successfully divided into several segmented regions,and then the centroid points of the segmented regions and the midpoints of the chords are connected to obtain the chordal axis transform skeleton line(axis).Through observation,it could be found that the segmented region around the contour of the pig represents the feature region to be identified.In order to distinguish the regions to be identified from other regions,a tree topology could be used to represent the hierarchical relationship between them,and these segmented regions around the contour of the pig are called the prominent protruding region.According to the characteristic of each feature region,a series of shape descriptors are set up to facilitate the follow-up classification and recognition by using support vector machines.With the characteristics of pigs in various forms,a detailed identification process was designed based on its topology.And the feature weighted support vector machine method is adopted,which improves the performance of the feature region recognition algorithm to some extent.In order to verify the performance of the proposed algorithm,1000 individual pig images with different positions and postures under random conditions were randomly selected as samples.The average recognition rate of the feature regions in the test set was 92.5%.In the analysis of the position accuracy of each feature region,the average position errors of the head,ear,nose,leg,and tail were 1.32%,0.75%,0.52%,0.42%,and 0.37%.The results show that the proposed algorithm has a good performance even when the pig raise head,bow head,expose leg or not.It provides new ideas for further behavior identification of pigs.