The Posture Recognition of Pigs Based on Zernike Moments and Support Vector Machines

Abstract

Walking postures of pigs usually reflect their health condition. When a pig lowers its head or walks very slowly for a long time, it is probably in bad health. And a pig walking quickly with its head up may mean it is irritated or frightened. These behaviours traditionally rely on manual observation. With the development of computer technology and digital image processing, machine vision gradually penetrated into various fields of agriculture production. In order to better monitor the behaviour of pigs, a new posture recognition method based on Zernike moments and support vector machines is proposed. First, the Otsu adaptive threshold segmentation is used to obtain the binary image. Then the contour of pigs is extracted by canny edge detection and morphological algorithms. Second, the Zernike moment feature parameters are extracted from the normalized binary contour images. Based on the above, the posture classifiers are designed according to support vector machine theory to recognize four kinds of behaviour postures of pigs, including walking normally, walking with head up, walking with head down, and lying. The experimental results show that the combination of Zernike moments and support vector machine makes the extracted features more sufficient and effective. And the posture classification accuracy of pigs reaches 95%.