Rapid Nondestructive Detection Method Of Mildew Rice Based On Hyperspectral Technology

Rice is a staple food of nearly half of the world’s population and plays an extremely important role in world food production.However,due to improper processing or storage methods,a large number of mildew rice occurs every year,resulting in a large amount of loss and waste,and even serious harm to people’s health.Ensuring the quality and safety of rice is crucial to ensuring food security in China and even the world.Therefore,it is very important to realize the rapid nondestructive detection of mildew rice.This study focuses on the rapid nondestructive detection of mildew rice,based on hyperspectral technology and combined with stoichiometry.The specific research was mainly carried out from three aspects: the identification of early mildew rice and healthy rice of the same variety and mixed varieties,the identification of different proportions of early mildew rice and healthy rice and early time analysis and identification of Aspergillus flavus infection in healthy rice.The specific research content is as follows:1.In the hyperspectral imaging system at 900-2,500 nm,the hyperspectral information of early mildew and healthy rice samples was obtained by line scanning reflection method,and the average spectral data of the region of interest were extracted.Three identification models including support vector machine(SVM),partial least squares discriminant analysis(PLS-DA)and principal component analysis combined with linear discriminant analysis(PCA-LDA)were established to identify mildewed rice from the same variety.Comparing and analyzing the classification results of the models,the three models could achieve ideal results for the identification of mildew rice,the PLS-DA model had the best identification effect,and the overall classification accuracy of the three kinds of rice was at least 95%.Above three classification models and linear discriminant analysis(LDA)were used to classify mixed varieties of mildew rice from healthy rice,analyzed of classification effect,then three spectral pretreatment methods including Savitzky-Golay smoothing method,standard normal variate(SNV)and standard normal transformation combined with trend algorithm(SNV-De-trending)were attempted to optimize the model classification and identification effect,finally,regression coefficient method(RC)and successive projection algorithm(SPA)were used to select feature bands,and SVM PLS-DA and PCA-LDA identification models based on feature wavelengths were established for mixed rice.The results showed that SPA-PLS-DA discrimination model had the best effect on the basis of spectrum processing by SNV-De-trending method,and the overall classification accuracy,sensitivity and specificity indexes were above 98%.At the same time,the good classification effect of various classification models also proved the feasibility of rapid and accurate classification of mildew rice and healthy rice.2.In the hyperspectral imaging system at 900-2,500 nm,the hyperspectral information of the samples mixed with mildewed rice and healthy rice was obtained by line scanning reflection method,and the average spectral data of the region of interest was extracted.SVM and PLS-DA models were established to identify each gradient sample data,the results showed that with the increasing proportion of mildewed rice mixed with healthy rice,the overall classification accuracy,sensitivity and specificity of SVM and PLS-DA models were mostly improved.SNV-De-trending method was used to optimize the identification model,and SPA algorithm was used to extract feature wavelengths for data dimension reduction.The classification model based on characteristic wavelength was established.When the proportion of mildew rice mixed with healthy rice reached 10%,the overall classification accuracy,sensitivity and specificity indexes of SVM model and PLS-DA model could reach more than 95%,showing good classification and identification effect.The results proved the feasibility of rapid nondestructive identification of mildew rice mixed with different proportions of healthy rice.3.In the hyperspectral imaging system at 900-2,500 nm,the hyperspectral image information of healthy rice samples inoculated with Aspergillus flavus at 12 h,24 h,36 h,48 h,60 h and the control group was obtained by line scanning reflection method,and the average spectral data of the region of interest was extracted,four identification models including SVM,LDA,PLS-DA and K-Nearest Neighbor(KNN)were established to study the identification of mildew rice inoculated with Aspergillus flavus at different times.The classification prediction results of the model were compared and analyzed.The results showed that when rice samples were infected with Aspergillus flavus for 24 h,the overall prediction accuracy of SNV-Detrending-PLS-DA model with the best classification prediction effect could reach 88.67%,which could basically realize the effective identification of mildew rice,with the growth of inoculation time.The overall classification accuracy,sensitivity and specificity of the four classification models were basically improved,when rice samples were infected with Aspergillus flavus for 60 h,the classification accuracy of each classification model for mildew rice could reach 100%.Then,Savitzky-Golay smoothing method,SNV and SNV-De-trending were used for spectrum preprocessing,and SPA algorithm was used to extract characteristic wavelength.The classification model based on characteristic wavelength was established,and the influence on the classification effect of the model was analyzed.In general,the results of this study have basically realized the early time analysis and effective identification of Aspergillus flavus infection in healthy rice.The results of the above study proves that the accurate and rapid nondestructive detection of mildew rice can be realized based on hyperspectral technology combined with appropriate stoichiometry methods.