Prediction Of Seed Germination Rate And Vigor Of Sugar Beet Based On HSI

Germination rate and vigor are important indicators to measure the quality of sugarbeet seeds.At present,the detection of germination rate and vigor is still dominated by traditional methods,which are highly specialized in operation and low in detection accuracy.The rapid and non-destructive prediction of seed germination rate and vigor before sowing is of great significance to the increase of sugarbeet yield and the development of related industries.This study combines hyperspectral imaging with machine learning to achieve accurate prediction of the germination rate and vigor of sugarbeet seeds.The main contents include:(1)Realized abnormal sample detection and dataset balance.The isolated forest was used to detect abnormal samples in the spectral data of sugarbeet seeds,and the samples with lower abnormal scores were regarded as abnormal samples and eliminated.Combining the synthetic minority oversampling technique and Tomek links to balance the sugarbeet seed dataset after removing the abnormal samples,a balanced dataset with the same number of positive and negative samples was obtained.(2)The prediction of germination and vigor of sugarbeet seeds was realized based on HSI.Spectral features,image features and fusion features of seeds were extracted to establish the prediction model for germination and vigor.In the full wavelength prediction,the prediction effect of the germination prediction model SNV+1D-Catboost and the vigor prediction model SNV+1D-SVM-RBF were better than other models,and the prediction accuracy of the test set reached 95.97% and 93.39%,respectively.The wavelengths extracted by the SPA were more effective when using characteristic wavelength prediction.The prediction accuracy of the germination prediction model SPA-Catboost and the vigor prediction model SPA-SVM-RBF were 92.48% and 91.78%,respectively,and the prediction effect was better.The texture features of the seed image were extracted using GLCM and fused with the extracted spectral features in the feature dimension.The prediction models were established based on image features and fusion features,respectively.The results showed that the performance of the image feature model was similar,but the prediction effect was not as good as the spectral feature model.The prediction performance of the model after feature fusion was significantly improved.Compared with the spectral feature model,the prediction accuracy of the germination and vigor model was improved by 1.04% and 1.17%,respectively.(3)Internal substances related to the germination and vigor of sugarbeet seeds were analyzed.The internal storage substances corresponding to the spectral features extracted by SPA were closely related to seed germination and vigor.Among them,gibberellin,glucose and starch had a great influence on the germination and vigor of sugarbeet seeds,which verifies the validity of the extracted spectral features.