Detection And Bruise Time Prediction Of Slight Bruise Of Peach Based On Hyperspectral Imaging

Slight bruise of fruit has a serious effect on the sale of fruit itself,deep processing and reputation of fruit sellers.For the slight bruised fruit is difficult to detect,affect the fruit taste and other problems.Based on the advantage of "Atlas in one" of hyperspectral,peach was token as the research object for this study.Spectral and image information of experimental samples were extracted respectively.The image segmentation algorithm and feature modeling algorithm were used to detect and analyze the slight bruise of fruit.And the feature modeling algorithm was used to make qualitative discriminant analysis on the time of bruising.The main research contents and conclusions of this paper were summarized as follows:(1)Qualitative identification of slight bruise of peach was realized based on hyperspectral image.Firstly,the principal component analysis(PCA)and minimum noise separation(MNF)were used to reduce the dimensionality of hyperspectral data.The MN4 image was selected as the analysis image.Finally,the fixed threshold method,the ostu method,and the improved watershed segmentation algorithm were used to segment the MN4 image.The final results showed that the improved watershed segmentation algorithm had the best effect.The accuracy of segmentation for minor bumps and non-strikes was 91.67% and95.00%,respectively.(2)Qualitative discriminant analysis of slight bruise of peach was carried out by using spectral feature and image feature modeling.Spectral and image features of the samples were extracted separately,and a qualitative discriminant discriminant model for minor peach bruises was established using partial least squares discrimination(PLS-DA)and least squares support vector machine(LS-SVM).The final results showed that the spectral feature model based on the LS-SVM algorithm RBF kernel function had the best prediction effect,and the model detection accuracy rate was 95.71%.(3)Hyperspectral images of bruised peach of 12 h,24h,36 h,and 48 h were collected respectively.Firstly,PCA algorithm was used to process the original data for obtaining the feature image,Then the spectral feature and the average gray value feature of the sample were extracted respectively.Finally,the spectral feature model,image feature model and mixed feature model of the peach peach injury time were established based on the LS-SVM algorithm.The results showed that the hybrid feature model based on the RBF kernel function had the best prediction effect,and the correct recognition rates of the samples at 12 h,24h,36 h,and 48 h were 83.33%,96.67%,100%,and 100%,respectively.The results show that Hyperspectral can better detect the slight bruises of fruits andestimate the time of bruises,which can provide a certain reference and basis for the external quality sorting of fruits,and has certain reference significance for fruit sales and deep processing enterprises.