code

To structure ensemble model ideology, semantic segmentation methods firstly apply to the training datasets, 20 results are generated with five semantic segmentation algorithms and four backbone architectures in stage 2 of Figure 1. Five semantic segmentation algorithms include CGNet, U-Net, DeepLabv3+, PSPNet, and DUC-HDC and the four backbones are plain networks, dilated convolutions, ResNet34, and DenseNet121. Therefore, a total of 20 base algorithms are obtained