Table 4.

Performance of these five algorithms: the fully convolutional networks (FCN), U-net, Encoder–decoder network (SegNet), the multi-scale fusion crack detection (MFCD) algorithm and the proposed U-CliqueNet.

Method	PA	MPA	MIoU (Mean Intersection over Union)	Precision	Recall	F1
FCN	0.9538	0.8825	0.8391	0.8269	0.7966	0.7854
U-net	0.9642	0.9059	0.8403	0.8457	0.8068	0.7967
SegNet	0.9385	0.8677	0.8050	0.7968	0.7495	0.7536
MFCD	0.9574	0.8850	0.7987	0.7937	0.8125	0.7808
U-CliqueNet	0.9661	0.9225	0.8696	0.8632	0.8028	0.8340