Since the total number of samples decreased significantly after processing, we en- hanced the data, randomly transformed all samples by rotating 90◦, 180◦ and 270◦ clock- wise, and obtained 7086 samples in total. Finally, we divided the samples into a training set, verification set ...