Elaboration of a new unsupervised method to extract body water

Abstract Maintenance of large groundwater resources is important for the development of ecosystems in the world. Remote sensing is now the main way to observe the world. Conventional body-water extraction methods are easy to implemente and use and work very well, but rely on a stable international scale, which reduces body-water extraction accuracy. In this paper, we present an original water body extraction unsupervised method for RGB satellite images . The idea is to splitt the three RGB component matrix of the imagea nd then treat only the green matrice beacause the green matrice contain all the water bodies with a great precision and only few non-water bodies regions. Our method is mainly based on mathematical morphology operators,First, we propose a simple and fast binary algorithm to segment as many real water bodies as possible from satellite images, this step was carried with the Hit or Miss Transform. The second step is to apply the Top Hat to refine the result of the segmentation. We tested the accuracy of our method using Sentinel-2Dataset. Compared to other methods, our method increases the segmentation quality and give great results that reached 98% for all index metrics used for remote sensing, the Error is equal to 0.7%. Keywords : Image processing, Mathematical morphology, RGB image, Classification, Body-water