Research on Carbon Foam Image Segmentation Based on Deep Learning

Table 3

The network training hyperparameters and experimental results.

Network	Epoch			IoU (%)	acc_carbon (%)
Deep-Res-MixAttention(Ours)	175	0.7543	0.6851	91.05	88.31
U-Net [22] (C1)	215	0.6785	0.7112	66.21	65.08
Res18-Unet(C2)	155	0.7435	0.7125	72.11	71.12
Res18-Unet+SEBlock [32] (C3)	198	0.7714	0.7781	75.25	73.15
Res18-Unet+CBAMBlock [33] (C4)	187	0.7321	0.7812	76.32	74.98
Res18-Unet+ECABlock [34] (C5)	190	0.7452	0.6993	78.25	75.45
Res50-Unet+SEBlock (C6)	185	0.7931	0.7551	82.81	84.32
Res50-Unet+CBAMBlock (C7)	195	0.7812	0.7852	85.62	86.51
AlexNet [35] (C8)	400	0.7125	0.7581	52.14	55.81
DenseNet [36] (C9)	45	0.7751	0.7332	74.21	72.51
Xception [37] (C10)	35	0.6325	0.7351	53.56	54.31
Mask RCNN [19] (C11)	40	0.7451	0.6999	68.48	66.82