Study On Grading Methods Of Dried Hami Jujube Based On Surface Fold

Dried Hami jujube has extremely high nutritional and medicinal value,and is favored by consumers.As demand of the society for red jujube continues to increase,the entrepreneurs' enthusiasm for processing red dates has also been increased.During the processing of dried Hami jujube,its surface wrinkles are an important grading standard for evaluating the external quality of dried Hami jujube.Therefore,it is necessary to sort the dried Hami jujube according to the surface folds of the dried Hami jujube to improve the attached value of the dried Hami jujube.In this paper,based on the previous theory and practical application.The machine vision acquisition platform was used to take images of dried Hami jujube for subsequent image processing and classification research of dried Hami jujube.The main research results of this paper are as follows:1?Using the single-scale two-dimensional discrete wavelet to decompose the image of the dried Hami jujube to obtain low-frequency signals and high-frequency signals.The low-frequency signal of the image can reflect the outline of the image,and the high-frequency signal can better reflect the noise and details mixed in the image.Therefore,the image reconstruction was performed by weakening the horizontal,vertical and diagonal signals of the high-frequency signal and then enhancing the low-frequency signal.The reconstructed dried Hami jujube images were used for the subsequent extraction of connected domains.In order to extract more rich texture information of the dried Hami jujube,this paper reconstructed the image by enhancing the horizontal,vertical and diagonal signals of high frequency signals and weakening the low frequency signals.2?In this paper,two previous algorithms were used to describe the density of connected domains to quantitatively describe the surface wrinkles of dried Hami jujube.In the first algorithm(A1),the centroid of the image was used as the origin of the coordinate system,and calculated the average distance from the centroid of each connected domain to the origin of the coordinate system.Calculated the distance from the center of each connected domain to the origin of the coordinate system,and then add them to average.In the second Algorithm 2(A2),the horizontal and vertical coordinates of the centroids of each connected domain in the image were averaged and used as the origin of the coordinate system,and then the distance from the centroid coordinates of each connected domain to the origin of the coordinate system was calculated.Finally,these distances that have been calculated were added and averaged as the feature value extracted by the second algorithm.Then,using the gray level co-occurrence matrix(GLCM)and color feature extraction algorithm to extract 6 texture features and 9 color features on the surface of the dried Hami jujube,respectively.In this paper,in order to betterevaluate the surface fold level of dried Hami jujube,SPSS software was used to analyze the correlation between the previously extracted image feature values(A1,A2 and 9 color features)and the 6 texture features extracted by GLCM.The results shown that except mH,mh and ms,all eigenvalues were significantly correlated with the maximum probability(Maxp),correlation(Corr),contrast(Con),energy(Asm)and consistency(Hom)in GLCM(Consistency and correlation are negatively correlated).Therefore,A1 or A2,Maxp,Corr,Con,Asm,Hom,mRGB,mSI,and mv were selected as the input feature of the model,which were used to grade the dried Hami jujube based on the surface folds.3?Two sets of data(A1 + Maxp,Corr,Con,Asm,Hom + mRGB,mSI,mv;A2 + Maxp,Corr,Con,Asm,Hom + mRGB,mSI,mv)were input into the Support Vector Machine(SVM)and the Extreme Learning machine(ELM),respectively.By comparing and analyzing the classification results,it was found that the SVM classification effect was better than the ELM classification model,and the selective fusion methods of A2 + Maxp,Corr,Con,Asm,Hom +mRGB,mSI,mv were superior to other feature fusion methods.In order to improve accuracy and avoid overfitting of the model.Therefore,this paper used the Zscore standardization method to pre-process the two sets of original data,and optimized the parameters of the SVM classification model by the grid search optimization method.The classification accuracy of the optimized model training set was 95%,the classification effect of the test set was 94.44%.4?Although the SVM model has a higher accuracy rate for dried Hami jujube classification based on surface folds,the process of image preprocessing and feature extraction is more complicated.Therefore,this paper used Convolutional Neural Network(CNN)to classify dried Hami jujube based on surface folds.Experimental results shown that the classification accuracy of CNN on dried Hami jujube according to surface folds was slightly higher than SVM classification model.In order to realize the classification of multiple dried Hami jujubes based on surface folds,this paper used Faster-RCNN to locate the images of dried Hami jujube and classify dried Hami jujubes based on surface folds.The experimental research shown that the positioning accuracy of this method for dried Hami jujube was better than previous scholars,and the classification accuracy had also been improved,reaching 97.30%.