A Novel Defect Detection Method for Ferrite Shield Surface Defects by Improved Faster R-CNN

Table 5

The model of SSD, YOLOv3, Retina-Net, and Cascade R-CNN are compared with the experimental results of the proposed model.

Models	Crazing	Pit	Impurity	Dirt	mAP	Time (ms)
SSD	0.626	0.817	0.645	0.539	0.657	39
YOLOv3	0.625	0.818	0.686	0.593	0.681	22
Retina-Net	0.689	0.779	0.674	0.659	0.700	76
Cascade R-CNN	0.745	0.759	0.691	0.718	0.728	82
YOLOv5	0.744	0.845	0.784	0.855	0.807	16
Proposed	0.853	0.926	0.790	0.674	0.810	51