A Recognition Method For The Head And Tail Of Adhesive Pigs Based On The Generalized Hough And Clustering

In order to analyze individual behavior in group-housed pigs,the complete contour and the main features of each pig must be obtained.The machine vision technology was used to segment the adhesive pigs and identify the head and tail of the individual,it is convenient to explore the individual analysis of pigs in future.Due to the complicated light in the piggery,more surface stains,the image quality is not high,the finite contrast adaptive histogram equalization(CLAHE)was used to enhance the image firstly in the paper.Then,the maximum inter-class variance method(Otsu)and morphological operation were used to obtain the ideal binary image.Thirdly,the canny operator was used to extract the binary contour of pigs.Finally,the images of adhesive pigs were selected by using the shape factor(SF)of the connected areas.Considering the effect of adhesion phenomenon on the identification of individual head and tail in group-housed pigs,and the traditional watershed algorithm is prone to over-segmented phenomenon in the process of segmentation,so the method of distance transformation and watershed algorithm is proposed to segment adhesive pigs in the paper.In the process of segmentation,the distance transformation was performed on the binary images to form splitting ridge map,and the map were superimposed with the original binary image.This method can accurately find the location of adhesion and reduce the phenomenon of over-segmentation.The experimental results show that the accuracy and complexity of the segmentation can meet the experimental requirements.Head and tail as the main feature of pig,it is very important to analyze individual behavior in group-housed pigs.Therefore,the generalized Hough transform and cluster analysis were used to identify the head and tail of pigs in this paper,because the generalized Hough transform can identify arbitrary shape target.On the basis of the segmentation of adhesive pigs,the rectangular coordinate system was used to intercept the symmetrical head-tail contour curve according to the second-order elliptical properties of the pig and the actual position in the image.Then the contour curves were sampled to form contour points.Each of contour points would generate corresponding K circles according to the principle of non-collinear three points to determine the only circle.All the circles were mapped from the image space to the parameter space by using the principle of generalized Hough transform.Because the shapes of the head and tail contours of pig are different,the head and tail were identified through dispersion of the center points.Experiments were carried out pigs in non-adhesive libraries and adhesive libraries,the results show that the average correct rate can reach 91.75% by using the recognition method in this paper.This study provides a new idea for behavior analysis of individuals in group-housed pigs.