High Resolution Remote Sensing Water Image Segmentation Based on Dual Branch Network

Abstract

A basic stage of hydrological research is to automatically extract water body information from high-resolution remote sensing images. Various methods based on deep learning convolutional neural networks have been proposed in recent studies to achieve segmentation. Such as FCN, PSPNet, Unet and so on. However, due to the complexity and multi-scale nature of high-resolution images, traditional segmentation networks cannot classify each pixel in the image well because they do not consider the spatial context information of the overall image, therefore, the results of segmentation often appear defects such as rough edges and inadequate water integrity.Based on the above reasons, this paper designs a two-way segmentation network DBAN based on spatial attention, which fully considers the detailed information and spatial context information of the image to refine the segmentation results.