Study On Quality Of Xinjiang Grape And Raisin Based On Machine Vision Technology

Fresh grapes and raisins are special agricultural products in xinjiang, but its processingtechnology after picking is backward, especially the quality inspection and grading, whichrestrict its development. Using machine vision technology to detect the external quality offresh grapes and raisins, to enhance the competitiveness of the industry has very importantsignificance. In this paper, machine vision technology was applied to detect the externalquality (diameter and color) of fresh grapes and raisins, and the corresponding classificationmodel was built for its external quality, It has very important significance to enhance thecompetitiveness of the fresh grapes and raisins industry, and lay the foundation of researchfor the equipment design for quality inspection by using machine vision technology on-line.The research content and results were as follows:(1) One object of this study is to forecast weight of Xinjiang thompson seedless grapespike by using machine vision technology. The RGB images of grape were acquired., R+Badd operation of G and B dual channel component was used. Gaussian low pass filteringmethod for filtering noise in images and gamma transform method for adjusting the imagegray level were used to enhance the visibility of the foreground from the background.Furthermore, the automatic threshold segmentation method was used to segment image. Thecorrosion and opening function of mathematical morphology were used to get best binaryimage. The geometrical characteristics of the target area of binary image were extracted. Forthe last, simple linear regression, multiple linear regression and partial least-squaresregression were used to predict weight of fresh grape. Results showed that partialleast-squares regression model by using the area, perimeter, length of long axis and shortaxis of the grape region of binary images is the best to predict the weight, and thecorrelation coefficient r2was96.91%.(2) In another objective, aiming at prediction and classification of the weight and sizefor single grape of Xinjiang thompson seedless and red grape. First of all, imagepre-processing and target area segmentation based on the method of maximal variancebetween-class were proceed in a different color feature space model. Secondly, mathematical morphology method was used to remove part of the stem and noise points in the binaryimage. The binary image was analysed for getting the geometrical characteristics of singlegrape. Finally, the method of simple linear regression and partial least-squares regressionwere respectively used to predict the weight and fruit size of the single grape. And quadraticdiscriminant analysis was used for single grape weight and size grading. Results shown thatthe partial least-squares regression model using short axis combined with fruit shape indexcharacteristics can effectively predict the weight and diameter of the single grape, the R2is0.98and0.945. Based on the characteristics of the combination of quadratic discriminantanalysis method can be used in single grape weight and size grading, and accuracy was morethan85%.(3) The third object of this study is to grade color features of xinjiang thompsonseedless raisins, HSI image extracting from RGB image, median filtering method, and OTSUmethod were used for image segmentation. Morphological open operation was used toremove the false target area in the binary image. The tonal cumulative gray histogram andcolor moments histogram were used to caculate the first, second and third moment of H, S, Icolor components. The results shown that color moments as characteristic values can be usedto establish the classification model based on BP neural network, the highest classificationaccuracy of grade color of xinjiang thompson seedless raisins was96.42%.