基于WOA-LSSVM的烟叶生态区模式识别研究OA

To achieve rapid and accurate discrimination of tobacco aroma styles,166 typical tobacco samples in eight ecological regions were selected for analysis of the conventional physical indices and 70 chemical components of tobacco leaves.Redundant features were eliminated through Spearman correlation analysis,and key variables were selected by SVM-RFE.Then,a WOA-LSSVM model was constructed and compared with the LSSVM and SVM models optimized by PSO and GA.The results showed that:(1)Chemical components such as Amadori compounds,organic acids,reducing sugars,and starch significantly contributed to the identification of tobacco leaf aroma styles,and physical indicators such as leaf surface density and equilibrium moisture content could also assist in the determination;(2)WOA-LSSVM outperformed other models in classification performance(training set recognition accuracy was 100%,F1-score was 1.00;test set accuracy was 95.9%,F1-score was 0.92)and computational efficiency(training time was 12.1s);(3)Based on the selected physical and chemical indicators of flue-cured tobacco leaves,using WOA-LSSVM to establish a model can achieve rapid and accurate identification of the tobacco leaf ecological region.This study can provide new ideas for tobacco production area traceability and aroma type evaluation.