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

Table 3

The classification accuracy of SE-ResNet-50 and SE-ResNet-101.

Item	Mix-fusion	SENet	SE-ResNet	SE-ResNet-50	SE-ResNet-101	Base
RCNN-DC	4.66	5.59	2.44	2.79	4.95	4.7
CNN	3.29	5.24	5.34	2.11	5.12	1.17
CBAM	4.36	4.5	2.16	4.96	4.87	3.81
MFLOPs	2.86	2.99	2.45	1.72	1.03	5.63
Straw	6.62	1.7	2.67	5.62	5.24	5.88
Multiply	4.92	5.77	5.52	1.61	5.33	3.59