Multilayer Feature Extraction of AGCN on Surface Defect Detection of Steel Plates

Table 2

Performance comparison (mIoU).

Model	White iron scale	Roll printing	Scratch	Scarring	Embroidery skin	Average
Faster R-CNN [8]	0.7962	0.7204	0.8386	0.7602	0.8982	0.8027
SegNet [39]	0.8250	0.6835	0.8532	0.8622	0.8824	0.8213
PSPNet [40]	0.8032	0.7282	0.8419	0.8358	0.9047	0.8228
YOLOv4 [14]	0.8068	0.7025	0.8793	0.8524	0.8856	0.8253
DeepLab+ [23]	0.8271	0.7139	0.8748	0.8437	0.8754	0.8270
RefineNet [41]	0.8265	0.7280	0.8786	0.8613	0.8410	0.8271
AGCN (ours)	0.8485	0.7188	0.9098	0.8762	0.9367	0.8580