OCNet-Based Water Body Extraction from Remote Sensing Images

Abstract

Water body extraction techniques from remotely sensed images are crucial in water resources distribution studies, climate change studies and other work. The traditional remote sensing water body extraction has the problems of low accuracy and being time-consuming and laborious, and the water body recognition technique based on deep learning is more efficient and accurate than the traditional threshold method; however, there is the problem that the basic model of semantic segmentation is not well-adapted to complex remote sensing images. Based on this, this study adopts an OCNet feature extraction network to modify the base model of semantic segmentation, and the resulting model achieves excellent performance on water body remote sensing images. Compared with the traditional water body extraction method and the base network, the OCNet modified model has obvious improvement, and is applicable to the extraction of water bodies in true-color remote sensing images such as high-score images and unmanned aerial vehicle remote sensing images. The results show that the model in this study can realize automatic and fast extraction of water bodies from remote sensing images, and the predicted water body image accuracy (ACC) can reach 85%. This study can realize fast and accurate extraction of water bodies, which is of great significance for water resources acquisition and flood disaster prediction.