Study On The Detection Method Of Live Pigs Based On Machine Vision

Live pig breeding industry occupies an important position in the economic construction in our country,in the process of pig breeding behavior monitoring,body quality and so on is of great significance to improve the pig meat.Traditional detection methods are mostly used methods such as scales,electronic scale,artificial measurement,some problems such as time-consuming,laborious,at the same time,the pigs in vivo in the process of the image capture is easily affected by light conditions,live pig detection based on image processing and machine vision technology has the advantages of non-contact,fast,and intelligent,to solve these problems,this study based on image analysis technology,data mining technology,in living pigs under complicated background image as the research object,to get the pig image filtering processing,live pigs target detection,morphological processing method will live pigs,and separation of complex natural background,and obtain a good separation effect,to the pig body quality testing make guidance,behavior analysis,etc.In this paper,the main research contents include:(1)For live pigs images in the process of obtaining will be affected by illumination,this paper uses the image preprocessing method of image contrast enhancement and light removal process.Through analysis,this paper argues that YCbCr color space can be a good characterization of light information,therefore,this article will get the live pig image in YCbCr space decomposition,Y channel can represent brightness information,so in the Y channel homomorphic filtering method was adopted to realize image pre-processing on the image.Results through the analysis of homomorphic filtering method,know the results of environmental illumination effect to remove the effect is not obvious,on the basis of this and the secondary wavelet decomposition and homomorphic filtering processing,make the image basic information at the lowest resolution layer.The experimental results show that,under the YCbCr space Y channel wavelet decomposition and homomorphic filtering was adopted to realize the reconstruction of the frequency domain,can reduce the live pig detection strong lighting effects,in the process of preparation for the subsequent live pig detection work.(2)In view of the live pig skin color difference image itself,based on machine vision method of pig skin detection and image segmentation.Based on the maximum likelihood classifier and iterative threshold method,namely,respectively,using the skin detection and image segmentation way will live pig separated from the complex background.In the maximum likelihood classifier adopted two RGB and HSV color space for detecting of live pigs,in a different color channel select training set and test set,build a supervised classification recognition model,for detecting living pigs.(3)For live pigs in their environment,and its color are more complex lead to the results of the live pig detection background noise,requires segmentation of image post-processing.This paper uses morphological operation was carried out on the live pig detection image de-noising processing,complete live pig contour is obtained through information.Finally,the confusion matrix is adopted to evaluate the processed image and analysis.Experimental results using maximum likelihood method and iterative threshold method,the average accuracy of 86.17%and 80.07%,respectively,correct error precision of 13.83%and 19.93%on average,show that the maximum likelihood method is adopted in leakage detection rate,accuracy and precision of error correct due to the iterative threshold method,this study has been basically achieved the detection requirements for living pigs.