Water image segmentation is often used in flood monitoring and water resources exploration,etc.Synthetic Aperture Radar(SAR)images often have shadows and interference spots,which leads to problems of poor connectivity and rough edges when using existing water segmentation methods.To make better use of the edge information of target to solve these problems,semantic segmentation and superpixel segmentation are utilized first to obtain the predicted results of water and superpixel clustering masks respectively.Then the hypothesis testing model is used to fuse the segmentation results of the two methods.Experiments are conducted on four semantic segmentation methods including U-Net,SegNet,DeepLabV3+and Superpixel Graph Convolutional Network(SGCN),and the segmentation accuracy of all the methods is improved.Among them,the improvement on U-Net is the greatest,with the mean Intersection over Union(mIoU)increased from 78.24％to 83.39％.Three different superpixel segmentation methods are also compared.The results show that the proposed method achieves the best performance on Linear Spectral Clustering(LSC).
维普
万方数据