Improved Faster R-CNN Based Surface Defect Detection Algorithm for Plates

Table 5

Comparison of the model proposed in this paper with other algorithms.

Algorithm	Accuracy (%)	Recall rate (%)	Training time (h)	Average inspection time/image (s)
Classification network + attention U-Net	85.27	83.36	6	0.32
Mask R-CNN	94.55	89.23	11	0.35
Cascade R-CNN	92.16	86.70	18	0.52
CBNet	94.50	89.58	19	0.57
DetectoRs	95.28	90.84	16	0.43
EfficientDet	89.74	85.41	8	0.23
YOLOv4	87.67	84.35	5	0.11
The model proposed in this paper	98.43	92.86	16	0.4