Research On The Identification Method Of Northern Japonica Rice Seed Varieties Based On Raman Spectroscopy

In recent years,the protection of agricultural germplasm resources has attracted much attention as a major livelihood issue.However,driven by interests,the seed market is in chaos,and there is no obvious difference between the appearance of seeds.Farmers cannot accurately and quickly distinguish them by naked eyes,and the phenomenon of purchasing is often not the one they choose.Therefore,It is of great significance for the accurate,efficient and rapid identification of rice seed varieties.At present,the identification method of rice seed is by original time-consuming,laborious manual method,gradually to the high accuracy,simple operation,the physical and chemical methods of transformation,in recent years,the Raman spectroscopy technology due to the advantages of rapid,efficient,convenient,development and application widely in the field of variety identification,studies have confirmed that Raman vibration spectrum of rice is the reflection of its internal composition and structure,which often have differences with different varieties.Therefore,this study attempts to use Raman spectroscopy combined with mathematical methods to achieve rapid identification of rice seed varieties.In this paper,33 kinds of japonica rice seeds mainly planted in Heilongjiang Province were taken as the research object.The spectral information of seeds was collected by Raman spectroscopy technology,and the identification model of PLSDA and SVM classification recognition algorithm based on different spectral data and the similarity analysis and discrimination model of correlation coefficient and cosine similarity were established.The main contents are as follows:(1)The acquisition of Raman spectra and the selection of spectral pretreatment methods.The experimental method and collection parameters were determined by collecting Raman spectrum data of japonica rice seeds.And further study the Savitzky-Golay(SG)smoothing,adaptive iterative reweighted penalized least square method(AIRPLS),first derivative(1-Der),second derivative(2-Der)and mean centralization(MC)and their combination of 13preprocessing methods in partial least squares discriminant analysis(PLSDA)And support vector machine(SVM)method to determine the best spectral pretreatment method.Through the study,in the spectral data of 400?3200cm^-1 band,the accuracy of AIRPLS+1-Der pretreatment data in the two types of models is 77.58%and 90.91%,respectively,which is better than other pretreatment methods.(2)Study on classification and recognition models of japonica rice seeds based on different spectral data and comparison of their effects.Due to the large amount of Raman spectral data and the long modeling time,PLSDA and SVM classification and discrimination models were established by using three different spectral information of the Raman spectra of japonica rice seeds,namely 400?3200cm^-1 band,400?1700cm^-1 band and eigenvalue.The discriminant effect and running time of the classification model under different modeling data conditions and different modeling methods are analyzed,so as to determine the better classification recognition model.After research,the less the amount of input data,the shorter the modeling time;Based on AIRPLS+1-Der pre-processed data,CARS features were extracted,and then the classification model established by SVM was used to identify 33 kinds of rice seeds.The identification accuracy reached 92.12%,indicating that Raman spectroscopy combined with mathematical methods had a better classification and discrimination effect on rice seeds.(3)Study on Similarity Analysis Method of Japonica Rice Seeds Based on Different Spectral Data and Comparison of Its Effect.The data of 400?3200cm^-1 band,400?1700cm^-1band and eigenvalue were also used to analyze the similarity of Raman spectral data of samples within the same variety and among different varieties,and the similarity analysis and discrimination model of japonica seed varieties based on correlation coefficient and cosine similarity were studied.The identification effect and running time of different models were compared and analyzed.The results show that the smaller the data amount is,the shorter the running time is.Correlation coefficient and cosine similarity method were used to establish the model in different data sets.The similarity of data of the same breed was better than the similarity between breeds.CARS feature extraction was conducted based on AIRPLS+1-Der pretreatment method.Finally,the discriminant model combined with cosine similarity algorithm has better discriminant effect on 33 kinds of rice seeds than other kinds.The above research results provide a theoretical basis for the efficient,convenient and rapid detection of rice seeds with multiple varieties,and also provide a new method and means for the research on the identification of seeds with multiple varieties.