Fig. 6

The classification tests of three different severity levels (1, 2, and 3) after training. Four groups were randomly set as test samples (n = 4 × 3 = 12). The four parameters were used individually and all as recognition features. Then they were input into the training model in turn. The mean P values were calculated by the outputs the training model (n = 4), which depicts the classification accuracy, ranging from 0 to 1. Larger mean P value corresponds to higher accuracy