A non-destructive testing method was studied to rapidly and accurately detect pesticide residues on mulberry leaves. Six groups of mulberry leaves were chosen as experimental samples, which contained pesticide residues of dichlorvos, chlorpyrifos, acephate, dimethoate and phoxim as the first to fifth groups, respectively, and the sixth group without pesticide residues was taken as control. Hyperspectral images of samples in 390~1.050nm were acquired by hyperspectral imaging devices. The region of interest from hyperspectral image was selected, and ten characteristic wavelengths, which were 452.51, 469.88, 517.28, 539.85, 578.92, 643.72, 727.24, 758.34, 785.67 and 819.67nm, were selected by the successive projections algorithm (SPA). Based on RBF kernel function of SVM and 10 fold crossvalidation methods, the detection models of pesticide residues on mulberry leaves were established. The impacts of three parameter optimization algorithms (grid search, genetic algorithm and particle swarm optimization) on the model performance were discussed. The results showed that performance of SVM model by using grid search was the optimal one, and its cross-validation accuracy was 63.89% and forecast accuracy was 78.33%. In order to further enhance the classification performance of the model, the adaptive algorithm (Adaboost) was introduced into the SVM model, and Ada—SVM algorithm was used to build classification model, which can detect pesticide residues on mulberry leaves and identify the kinds of pesticide residues. The results showed that the prediction accuracy of Ada—SVM model reached 97.78%, which was increased by 19.45% compared with the original SVM model. Therefore, hyperspectral imaging technology combined with Ada—SVM algorithm can accurately identify the pesticide residues on mulberry leaves. 