一种基于对抗学习的高分辨率遥感影像语义分割无监督域自适应方法

潘俊杰, 慎利, 鄢薪, 聂欣, 董宽林

An adversarial learning-based unsupervised domain adaptation method for semantic segmentation of high-resolution remote sensing images

PAN Junjie, SHEN Li, YAN Xin, NIE Xin, DONG Kuanlin

表2 Potsdam→Vaihingen对比实验的结果评价

Tab.2 Evaluation of the results of the Potsdam → Vaihingen comparative experiment

模型 mIOU IOU 其他类 汽车 树木 低矮植被 建筑 道路 Deeplabv2 0.264 7 0.066 3 0.074 5 0.186 5 0.244 3 0.488 6 0.527 9 AdaptSegNet 0.423 1 0.075 2 0.263 4 0.457 8 0.401 1 0.720 1 0.620 9 CLAN 0.410 1 0.084 7 0.164 0 0.544 1 0.274 1 0.773 0 0.619 9 ADVENT 0.434 8 0.168 6 0.221 8 0.510 7 0.316 9 0.768 2 0.622 4 Metacorrection 0.440 4 0.102 8 0.249 5 0.517 1 0.400 1 0.744 8 0.628 1 OA-GAL 0.474 8 0.114 8 0.219 5 0.573 2 0.435 5 0.818 2 0.687 4