Water body classification from high-resolution optical remote sensing imagery: Achievements and perspectives

Water body classification from high-resolution optical remote sensing (RS) images, aiming at classifying whether each pixel of the image is water or not, has become a hot issue in the area of RS and has extensive practical applications in a variety of fields. Numerous existing methods have drawn broad attention and achieved remarkable advancements, meanwhile, serious challenges and potential opportunities also exist, which deserves in thinking and discussing deeply. By taking into account the comprehensive survey is still lacking, through the compilation of approximately 200 papers, this paper summarizes and analyzes the achievements, and discusses the perspectives of future research directions. Specifically, we first analyze 5 challenges according to the characteristics of water bodies in high-resolution optical RS imagery, and 5 corresponding significant opportunities combined with advanced deep learning techniques are discussed to respond mentioned challenges. Then, we divide the existing methods into several groups in light of their core ideas and introduce them chiefly. In addition, some practical applications and publicly open benchmarks are listed intuitively. 10 and 9 representative methods are implemented on two widely used datasets to assess their performance, respectively. To facilitate the qualitative and quantitative comparison in the research avenue, the two benchmarks employed in the comparative experiments and links to other relevant datasets and open-source codes will be summarized and released in https://github.com/Jack-bo1220/Benchmarks-for-Water-Body-Extraction-from-HRORS-Imagery. Finally, we discuss a range of promising research directions to provide some references and inspiration for the following research. The studies of our paper, including the existing methods, challenges, opportunities, derived applications, and future research directions, provide a fuller understanding of water body classification from high-resolution optical remote sensing imagery.