Study On The Detection Model Of Lotus Seed Amylose Content Based On Hyperspectral Imaging Technology

The content of amylose in fresh lotus directly affects the subsequent processing quality and taste.The traditional amylose content detection method is time-consuming,and it is difficult to achieve batch detection.Hyperspectral imaging technology can simultaneously obtain the spectral and image information of the sample,which has been widely used in the field of nondestructive testing of agricultural products.In this study,amylose content was detected based on hyperspectral imaging.The main results are as follows:(1)Comparing with Different spectral preprocessing methods were used to preprocessing based on the spectra extracted from the region of interest.Comparing with the first derivative,second derivative,MSC,SG based PLSR model,it shows that MSC is the best preprocessing method.The correlation coefficient of the calibration set was 0.803,the root mean square error of the calibration set was 2.072 mg/g,the correlation coefficient of the prediction set was 0.836,the root mean square error of the prediction set was 1.855 mg/g,and the relative analysis deviation was 1.623.(2)For the MSC preprocessed spectra,the best feature wavelength selection method is preferred.In this study,feature wavelengths were extracted by regression coefficient,continuous projection algorithm and competitive adaptive weighting algorithm.The results show that the feature wavelengths for the three characteristic wavelengths are different,and the number of preferred wavelengths for SPA is the least,which are449.5 nm,464.1 nm,500.6 nm,582.7 nm,644.3 nm,756.1 nm,812.4 nm,872.3 nm,and 977.6 nm,a total of 9 characteristic bands.(3)The preferred different wavelengths were used as the input of different models,and different fresh lotus amylose detection methods were compared.In this paper,PLSR,PCR,SVM and LS-SVM were used respectively,and the results showed that MSC-SPA-LS-SVM detected the best results,The correlation coefficient of the calibration set of the MSC-SPA-LS-SVM model is 0.855,the root mean square error of the calibration set is 1.631 mg/g,the correlation coefficient of the prediction set is 0.841,the root square error of the prediction set is 1.697 mg/g,and the relative analysis deviation is 2.051.(4)Eight kinds of image texture features were extracted based on the gray-scale co-occurrence matrix,and the results of different detection models were established by using texture features and combining texture features with visible and near-infrared spectra,and established a fresh lotus amylose detection method based on hyperspectral imaging technology.The results show that the use of image texture features in this study is not superior to the model that only uses the spectrum of the region of interest.The detection results of the LS-SVM model established by combining texture features and visible and near-infrared spectra are the correlation coefficient of the modeling set is 0.847,the root mean square error of the modeling set is 1.876 mg/g,the correlation coefficient of the prediction set is 0.849,the root mean square error of the prediction set is 1.743 mg/g,and the relative analysis deviation is 1.897.The results were slightly reduced compared by the MSC-SPA-LS-SVM established using visible near-infrared spectra.