Optimal extraction of reservoir water body from remote sensing images based on iterative inter-class variance maximization method

To delineate the water body from remote sensing images using an empirical threshold, it requires many trials. However, it is difficult to make an objective determination on the selection of segmentation threshold between the water body and other features. Therefore, based on the iterative inter-class variance maximization method, this study proposed an iterative inter-class variance maximization method for the optimal extraction of the water bodies of reservoirs. We derived the preliminary water body information from the GF-1 satellite images using normalized difference water index(NDWI). Then we optimally distinguished the water and non-water bodies in the buffer zone, that is established by the morphological dilate algorithm, using the adaptive thresholds determined by the iterative inter-class variance maximization method. The results show that the method can effectively eliminate the influence of buildings and accurately obtain the characteristics of the water body information in the reservoir area during different periods. Compared with the results of iterative inter-class variance maximization method, the overall accuracy, Kappa coefficient, and comprehensive accuracy of the new method were improved by 9.36%, 24.09%, and 10.42%, respectively. © 2021, Editorial Board of Water Resources Protection. All rights reserved.