The Pig Posture Recognition Research Based On Feature Extraction And Feature Optimization

The pig posture recognition is the important preparation for pig behavioranalysis and digital cultivation. In order to get pig posture accurately andquickly, feature optimization is proposed to recognize pig posture. This paperreplaces manual watching with intelligent monitoring, and recognizes the pigposture by combining image processing with pattern recognition. This methodnot only protects the workers’ health and reduces the cost, but also discovers theabnormal pig posture in time, so as to take corresponding measures.First of all, this paper introduces the research development of pig posturerecognition and the feature optimization in the world. Considering the lackfeature optimization when describing the characteristics of a pig in the presentstage, this main content of the paper is as follows:Firstly, analyzes the differences between the object and the surroundingbackground, this paper presents a pig object extraction algorithm by combiningcolor and texture information. The Grey Level Dependence Matrix (GLCM) has been used to extract the texture features and the attribute reduction algorithm ofrough set has been applied to determine the optimal texture feature. The papercharacterizes each pixel by combining color and texture features, and uses theMean Shift algorithm to finish clustering. Then morphological image processingand the median filter algorithm have been used to eliminate small cavities andnoises. The pig outline is extracted more completely from complex images ofthe pig.Secondly, the advantages and disadvantages of several edge detectionmethods have been analyzed and compared. The Canny edge detection methodhas been chose to extract the outline of the pig. The alternative morphologyfeature set of pig is established based on the binary image that is circularity F1,rectangularity F2, elongation F3, ratio of the height of centroid and height F4andleft-right ratio of centroid F5. The edge moments feature set of pig is establishedbased on the outer contours. For standing and lying postures, and horizontal,lowered and raised heads postures, the two sets are selected through the attributereduction thought of rough set theory.Finally, this paper expounds concrete ideas of template matching method,and establishes sample libraries of all the postures. By comparing the posturerecognition of the two groups before and after reduction feature set, the featurecombination is optimized based on the principle of features selection. Theresults of the experiments show that, different optimal combination is selectedaccording to various posture recognition objects due to the different decision goal. The postures recognition of standing and lying achieves higher accuracyand lower complexity with the optimization result of feature set {F1,F2} and therecognition accuracy of posture recognition of horizontal, and the postures oflowered and raised heads achieves a higher recognition accuracy withoptimization result of feature set {F1,F2,F3,F4,F5}.