Quality Inspection And Variety Discrimination Of Nectarine Based On Hyperspectral Imaging Technology

Nectarine originated in China and it was the variety of the ordinary peach. Because of its strong adaptability to soil and climate, it had been widely cultivated. Nectarine was favored by people because of its bright color and sweet flavor. In this study, nectarine was chosen as the research object. The external defects and the internal quality of nectarine had been detected based on hyperspectral imaging technology, also many kinds of nectarine had been analyzed and discriminated at the same time..The results and conclusions of this study were as follow:(1) The external defect of "Zhongyou 9" nectarine had been detected from the spectral angle based on hyperspectral imaging technology. Using the spectral data of the visible/near infrared spectra (420 to1000nm), the classification model of PLS (partial least squares), LS-SVM (least square support vector machine) and ELM (extreme learning machine) were established in the full band, the principle components and the characteristic wavelength, respectively. The models were used to discriminate the normal, crack, peel spots, malformation and hidden damage nectarine. The results indicated that the optimal model was the LS-SVM model established in the full band, and its discrimination accuracy was 96.67%. The ELM model was more preferable than PLS and LS-SVM in the band of the principle components, and its discrimination accuracy was 90.00%. The LS-SVM model were the best model in the band of characteristic wavelength obtained by PLSR (partial least squares regression) method and SPA (successive projections algorithm) method, and the discrimination accuracy were 95.00% and 94.17%, respectively.(2) The external defect of "Zhongyou 9" nectarine had been detected from the image identification angle based on hyperspectral imaging technology. The principal component analysis method had been used to analyze the image of the defect sample to obtain the corresponding principal component images (PC images).Through the analysis and comparison of different principal components, the image of PC-4, PC-3, PC-1 and PC-5 were used to segment the defects region of the normal, crack, peel spots, malformation and hidden damage nectarine, respectively. Because the sample of nectarine had the highest reflectivity in 876nm, Therefore, the PC image had been done mask processing by using the gray image of the sample in 876nm. The edge image of the sample had been extracted from the image after mask processing by using Sobel operator. The defect image of the sample had been extracted by using region growing algorithm. The results indicated that the discrimination accuracy was 92.50%.(3) The external defect of Zhongyou 9" nectarine had been detected from the texture angle based on hyperspectral imaging technology. The interest region of sample had been extracted by using ENVI software, and the size of the extracted region was 50 by 50 pixels. The six texture index (mean, contrast, correlation, energy, homogeneity and entropy) had been chosen to compare different model. The results indicated that the optimal model was the LS-SVM model, and the discrimination accuracy was 88.33%.(4) The internal quality of nectarine had been detected based on Hyperspectral imaging technology. Flesh hardness, Soluble solids content, Organic acid and the content of Vitamin C had been analyzed in two different spectral band (420-1000nm..900-1700nm). The optimal model had been obtained by comparing the three models (PLS, LS-SVM and ELM) established in different spectral band and different pretreatment method. The results indicated that the PLS model established by using SPA method in Vis/Near-infrared spectral band was the optimal model of the flesh hardness. The R2P and RMSEP were 0.8645 and 2.7363. The ELM model established by using PCA (principle component analysis) method in Vis/Near-infrared spectral band was the optimal model of Soluble solids content. The R2P and RMSEP were 0.8323 and 0.0622. The LS-SVM model established by using PLSR method in Near-infrared spectral band was the optimal model of the content of Organic acid. The R2P and RMSEP were 0.7481 and 0.0250. The LS-SVM model established by using SPA method in Near-infrared spectral band was the optimal model of the content of Vitamin C. The R2P and RMSEP were 0.8029 and 0.1183.(5) Spectral information and Texture information of the image data had been fused based on hyperspectral imaging technology. And the detection of the external defects and soluble solids content of the sample was realized at the same time. Using the fusion of PCA and image texture as the input variables, the models (PLS, LS-SVM and ELM) were established to discriminate the external defect and the content of Soluble solids. The results indicated that the LS-SVM model was the optimal model. And the discrimination accuracy of external defect was 93.33% and The R2P and RMSEP of internal quality were 0.8747 and 0.9101.(6) The 4 different kind of nectarine had been qualitative discrimination from the spectral angle based on sub-band hyperspectral imaging technology. The discriminant models (PLS, LS-SVM and ELM) were established on the full spectral band, the principle components and the characteristic wavelength in both Vis/Near-infrared band and Near-infrared band, respectively. The result indicated that the model of Vis/Near-infrared band was more preferable than Near-infrared band and PLS model was more preferable than others. The discrimination accuracy of the full band model, the principle components model, the PLSR method and the SPA method were 96.43%,92.37%,94.92% and 93.33%, respectively.(7) The 4 different kind of nectarine had been qualitative discrimination from the image angle based on hyperspectral imaging technology. The texture of image had been extracted by GLCM in Vis/Near-infrared band. Then established three discriminant models (PLS, LS-SVM and ELM) to discriminant the 4 kind of nectarine. The result indicated that the PLS model was the optimal model, and the discrimination accuracy was 81.49%.(8) The 4 different kind of nectarine had been qualitative discrimination from the fusion of image and spectral angle based on hyperspectral imaging technology. The principle component and the characteristic wavelength that extracted by using PLSR and SPA had been fused with the texture of image to establish the discriminant models (PLS, LS-SVM and ELM). The result indicated that the LS-SVM model which fused the texture of image and characteristic wavelength extracted by PLSR was the optimal model, and the discrimination accuracy was 92.46%.