全局信息提取与重建的遥感图像语义分割网络

梁龙学,贺成龙,吴小所,闫浩文

Remote sensing image semantic segmentation network based on global information extraction and reconstruction

Longxue LIANG,Chenglong HE,Xiaosuo WU,Haowen YAN

表 4 在Vaihingen测试集上与先进的高精度网络定量比较结果

Tab.4 Quantitative comparison with advanced high-precision network on Vaihingen test set

方法 骨干 F1/% F1mean/% OA/% mIoU/% 不可渗透 建筑 植被 树 车 杂物 DANet(2019) Resnet18 90.3 93.9 82.5 88.3 75.8 54.1 86.2 88.8 76.2 ABCNet(2021) Resnet18 90.6 93.0 81.5 89.6 84.2 38.1 87.8 88.7 78.5 MANet(2022) Resnet18 92.0 94.5 83.5 89.4 88.0 50.9 89.5 90.0 81.1 BANet(2021) ResT-Lite 92.4 95.1 83.8 89.8 89.0 54.5 90.0 90.5 82.1 MAResUNet(2021) Resnet18 92.2 95.1 84.3 90.0 88.5 50.9 90.0 90.5 82.0 UnetFormer(2022) Resnet18 92.7 95.4 84.4 90.1 89.7 57.1 90.5 90.8 82.8 DCswin(2022) Swin-tiny 92.5 95.5 84.7 90.2 88.8 44.9 90.4 90.8 82.6 MAGIFormer MSCAN-tiny 92.7 95.3 84.7 90.3 89.8 53.7 90.6 90.9 82.9