Table 3.

The prediction error of the classification module, which is trained by Kullback–Leibler divergence loss with/without data augmentation.

	Accuracy_Severity	Accuracy_Number	RMSE_Count
Case 1	99.31%	84.60%	2.74
Case 2	99.17%	84.17%	2.17
F-RCNN		73.97%	3.39
YOLOv3		63.70%	3.37
Wu et al.		84.11%	2.33