Pig Posture Recognition Based On Object Features And Abnormal Behavior Evaluation

With the large-scale and automated development of the aquaculture industry,relying on manual observation is less effective and increases labor costs in monitoring the abnormal behavior and health status of pig.Therefore,in order to improve the efficiency of monitoring,using machine vision monitoring instead of manual monitoring in intensive pig farms has been gradually emerged.After the analysis of livestock and poultry behavior detection based on machine vision and abnormal analysis of pig,this paper proposes a method for pig posture recognition based on object features and abnormal behavior analysisis.Obtaining the video image of pig farm based on the video monitoring system,the object detection and posture recognition of the pig is researched by the machine vision technology.In combination with the statistical information on the daily behavior of pig,the abnormal evaluation system is established.The main research works of this paper are as follows:(1)Based on theoretical analysis of the bimodal histogram method,improved OTSU algorithm,and color-based object detection algorithm of pig,the experimental results of using color feature to extract pig object are satisfactory combined with experimental comparison and analysis.The pigobject detection image is optimized through filtering and morphological processing.In order to further improve the clarity and integrity of pig object detection image,in view of the contour of pig object detected by the mean-shift segmentation method is not accurate enough,it is susceptible to noise interference,and there are holes inside the object.This paper proposes a method of combining the super pixel information and the mean-shift algorithm,the pig object and the background can be clearly divided using the segmentation algorithm of SLIC super pixel,so fusing the superpixel segmentation information of pig can improve contour and fill hole of pig binary image detected by mean-shift.(2)This paper selects the common posture of pig to be studied and establishes the corresponding image sample database.The contour of pig object is extracted by using Canny edge detection operator.By comprehensively considering the extraction algorithm of image feature and the characteristic of pig posture,the object features set of pig posture is established,including twelve features,that is,circularity,rectanglarity and Hu invariant moments.The feature selection of the feature set based on samples is judged by within-class and among-class distance criteria.In order to realize the accurate identification of different pig postures,a decision tree support vector machine(DT-SVM)is constructed to identify the posture of pig hierarchically based on the logic of the decision tree and the support vector machine algorithm.The parameters of RBF kernel function are optimized by the method of grid search and cross validation.Compared with the experiment results of 1-V-1 SVM,the DT-SVM has higher accuracy in identifying the pig posture.(3)The posture image of pig is annotated by using the established DT-SVM classification model.Two key behaviors of rest sleeping and daily activities are selected as evaluation items based on the behavioral anchored rating scale.The statistics for evaluation items of pig key behavior are conducted.According to the differences between the specific behavior data and the overall statistics database,an abnormal evaluation system of pig is established to lay a theoretical foundation for the automatic monitoring the abnormal behavior of pig,which is conducive to raising the automation level of pig aquaculture industry.