Multiscale Road Extraction in Remote Sensing Images

Table 1

Comparison with U-Net, LinkNet34, DeepLabv3+, and D-LinkNet on the test datasets.

Model	Pixel accuracy	mIoU
U-Net	0.978	0.793
LinkNet	0.981	0.810
DeepLabv3+	0.983	0.831
D-LinkNet	0.982	0.821
Ours	0.983	0.823