Casting Defect Detection and Classification of Convolutional Neural Network Based on Recursive Attention Model

Table 4

The impact of different modules on network performance.

Item	FLOPs	XLD	Defect detection	Image processing	PyTorch	ResNet-50
ShuffleNetV2	3.02	4.81	5.42	5.78	2.4	6
Mix-fusion	1.35	2.93	3.31	5.39	5.01	2.69
Loss	4.73	4.3	1.98	3.32	3.82	3.98
ShuffleNet	1.5	1.3	1.24	3.19	1.76	6.82
Fourier	2.41	5.52	3.09	5.84	5.68	4.72
Casting	1.98	6.39	1.28	4.75	6.67	2.84