Study On Analysis Of Pepper Phytophthora Blight And Quality Based On Hyperspectral Imaging

The chili pepper（Capsicum annuum L.）is an important vegetable around the world,particularly in Asian and South American countries.In order to promote the development of modern pepper industry,all pepper breeders are working on finding and cultivating new chili pepper cultivars with early maturity,high yield,stress resistance,disease resistance and high quality now.The realization of these objectives depends first on the study of pepper germplasm resources,among which disease resistance and quality research have always been the focus of pepper breeding.However,the traditional methods are often time-consuming,expensive and cumbersome,so it is very important to find a faster and more efficient method.Hyperspectral imaging technology has broad application prospects in crop disease and quality detection due to its accurate,rapid and non-destructive advantages.The analysis of pepper phytophthora blight and quality based on hyperspectral imaging technology were explored in this paper,which provided theoretical basis and technical support for the subsequent precision breeding and screening of high-throughput germplasm resources of chili pepper.The main findings are as follows:（1）Six chili pepper cultivars with different resistance to pepper phytophthora blight were selected and a robust early diagnosis model of pepper phytophthora blight was established based on hyperspectral imaging technique.Five optimal wavelengths（550 nm,670 nm,722 nm,760 nm and 800 nm）were obtained by genetic algorithm-partial least squares（GA-PLS）combined with correlation analysis.Eight optimal vegetation indices were selected by Pearson correlation analysis,namely disease water stress index（DWSI）,enhanced vegetation index（EVI）,green index（GI）,greenness vegetation index（GVI）,nitrogen reflection index（NRI）,photochemical vegetation index（PVI）,red-edge vegetation stress index（RVSI）and triangular vegetation index（TVI）.The classification models of healthy and infected asymptomatic chili pepper samples were established by using support vector machine（SVM）algorithm based on full spectrum,all vegetation indices,optimal wavelengths and optimal vegetation indices.Finally,the optimal model is a classification model based on the optimal wavelengths.On the 5 days after inoculation,the overall accuracy reaches 91.25%.The results showed that chili peppers of healthy and infected asymptomatic could be accurately identified by hyperspectral imaging on the 5 days after inoculation.（2）Different infection levels of chili pepper samples were selected and a robust classification model of pepper phytophthora blight was established based on hyperspectral imaging technique.The classification model of pepper phytophthora blight infection level was established by using extreme learning machine（ELM）,k-nearest discriminant（KNN）and SVM algorithm based on full spectra,five optimal wavelengths and eight optimal vegetation indices.The classification results of all the models based on the optimal wavelengths is equal to that of the models based on the optimal vegetation indices,and the overall accuracy reaches 98.00%.The results show that the five optimal wavelengths and eight optimal vegetation indices selected in this study are universal for the identification of pepper phytophthora blight,namely it is not only suitable for early diagnosis model of pepper phytophthora blight,but also suitable for classification model of pepper phytophthora blight infection level.（3）Different cultivars of chili peppers at different ripening stages were selected and a comprehensive and reliable model of pepper quality identification based on hyperspectral imaging technology and visual analysis were established.The variation rules of capsaicinoids concentrations for different chili pepper cultivars at four ripening stages were studied.The capsaicinoids concentrations in the low-pungency chili peppers began to accumulate in the initial stage until the highest concentrations in the fully mature fruit,while the high-pungency chili peppers grew and accumulated from the beginning,but there was a slight decline in green maturity.Three different variable selection methods with successive projections algorithm（SPA）,competitive adaptive reweighted sampling（CARS）and GA-PLS were performed to select the optimal wavelengths.Partial least squares（PLS）,least-squares support vector machine（LS-SVM）,radial basis function neural network（RBFNN）and ELM models were then developed to predict the capsaicinoid concentrations and the water content.The results show that the ELM models combined with the SPA method yielded the best prediction performances for the capsaicin and dihydrocapsaicin concentrations,and the water content,with the highest correlation coefficients of prediction（R_P）of 0.83,0.80and 0.93,and the root-mean-square error of prediction（RMSEP）of 0.0156 g kg^-1,0.0168 g kg^-11 and 0.58%,respectively.The RBFNN model combined with the SPA method yielded the best classification performances with the overall accuracy of98.00%,the Distribution maps of capsaicin and dihydrocapsaicin concentrations for intact and cut chili peppers were obtained.These results indicate that a fast and non-destructive on-line identification method for pepper quality can be established based on hyperspectral imaging technology,which will be helpful for pepper industry.