基于Unet网络多任务学习的遥感图像建筑地物语义分割

刘尚旺, 崔智勇, 李道义

Multi-task learning for building object semantic segmentation of remote sensing image based on Unet network

LIU Shangwang, CUI Zhiyong, LI Daoyi

表2 不同方法的实验结果

Tab.2 Experimental results of different methods(%)

城市 FCN+MLP VGG16 VGG16+边界预测 ResNet50 本文方法 IoU Acc IoU Acc IoU Acc IoU Acc IoU Acc Austin 61.20 94.20 70.66 95.28 72.81 95.82 72.38 95.79 74.41 96.09 Chicago 61.30 90.43 66.37 91.44 67.38 91.92 66.12 91.50 67.76 92.02 Kitsap Co. 51.50 98.92 57.55 98.19 57.54 98.90 58.68 98.95 60.19 98.63 West Tyrol 57.95 96.66 67.82 95.35 67.18 97.01 67.32 97.07 69.09 97.74 Vienna 72.13 91.87 77.01 93.28 77.19 93.31 76.86 93.21 78.21 93.63 均值 64.67 94.42 69.82 94.71 71.61 95.39 71.08 95.30 72.53 96.10