ILCS: An Improved Lightweight Convolution Structure and Mixed Interactive Attention for Steel Surface Defect Classification

Table 3

Params (M) and FLOPs (G) of module in the NEU dataset.

	Params (M)	FLOPs (G)
ResNet	25.56	65.78
ResNet + MIAM	25.57	65.79
MobileNet	2.23	4.70
MobileNet + MIAM	2.24	4.71
EffNet	2.21	8.35
EffNet + MIAM	2.22	8.36
ILCS (ours)	2.24	4.26

Bold values are the ablation experimental results of our attention block and the Params (M) and FLOPs (G) of our ILC in the dataset. Our method only increases a small number of parameters, but greatly improves the accuracy.