Study On Soybean Seed Variety Classification Based On Spectral Technology

The main producing area of soybean in China is northeast.Due to the wide planting area and variety of soybean in northeast,the varieties suitable for planting soybean seeds in different regions are different.However,driven by interests,soybean seed market is chaotic.Farmers can not accurately and quickly identify soybean seed varieties with the naked eye,often appear the purchase of non-selected phenomenon.Therefore,the rapid and accurate classification of soybean seed varieties is of great significance.However,the traditional sensory detection methods have large workload and strong subjectivity.Chemical detection methods are accurate but require a long time.In order to improve this situation,this paper carried out the classification of soybean seed varieties based on spectral technology.The principle and characteristics of hyperspectral imaging technology are analyzed,and the processing and data analysis methods of hyperspectral data are further studied.On this basis,hyperspectral images of different soybean varieties in 392.38 nm～1011.01 nm were obtained by hyperspectral imaging technology,and the region of interest（ROI）data were extracted to obtain the reflectance spectrum curve of soybean samples.After S-G smoothing,extreme learning machine（ELM）and random forest（RF）models were used to model and analyze the full-spectrum reflectance spectra of different varieties of soybeans.The classification accuracy of ELM and RF was 65.67%and 90.00%,respectively.The model prediction time is 13 s and 158 s.According to the difference of reflection spectrum curve of different soybean varieties,455.54 nm,479.3 nm,604.04 nm,657.46 nm,705.72 nm,856.89 nm,918.07 nm and 953.54 nm were selected as the characteristic bands of reflection spectrum curve,and the same classification method was used for the characteristic bands.The classification accuracy of ELM and RF was 78.22%and 98.89%,respectively.After timing,the prediction time of the model is 11 s and 12 s.The classification methods were compared,and the hyperspectral data processing and analysis method with high classification accuracy and fast speed was obtained.The principle analysis,theoretical calculation and visual classification of Raman spectroscopy were carried out.The spatial structure of oleic acid and linoleic acid molecules in soybean varieties was constructed by density functional theory.The molecular structure was optimized by B3LYP/6-31+G（d,p）basis set,and the theoretical Raman spectroscopy was calculated.Raman spectroscopy detection experiment was carried out to measure the Raman spectra of oleic acid and linoleic acid and six varieties of soybean.The theoretical Raman calculations of oleic acid and linoleic acid were compared with the experimental Raman spectra,and the Raman spectra at 1281 cm^-1,1445 cm^-1,1662 cm^-1and 2904 cm^-1were identified,which were basically consistent with the Raman peaks measured by six kinds of soybeans.The four Raman peaks were used as the Raman characteristic peaks of soybean,and the soybean seeds were classified and identified by principal component analysis（PCA）and discriminant analysis（LDA）.In the PCA score map,soybeans were clustered by varieties.The LDA model training set and test set were used to discriminate,and the correct rates were 93.52%and 90.00%,respectively.The effective classification method of soybean varieties based on Raman spectroscopy was obtained.The research methods and results of this paper not only provide a reference for the classification of soybean varieties,but also provide a certain reference for the classification of substances in the field of spectrum.