Taking the GF1 remote sensing image as the data source and the Dongfanghong Reservoir in Tunxi Watershed, Huangshan City, Anhui Province as the research object, five kinds of water and land pixel measurement methods, including single band threshold method, two band difference method, band ratio method, normalized difference water index (NDWI) method, and normalized difference vegetation index (NDVI) method, were used to extract water body of the Dongfanghong Reservoir, respectively, using the average method and the maximum interclass variance iteration method, to explore the improvement effect of the maximum interclass variance iteration method on water body extraction variables. On this basis, an improved maximum interclass variance joint water extraction method was proposed, and the water extraction effects before and after the improvement were compared. The results show that the improved water body extraction method can effectively reduce the noise generated in image extraction and improve the accuracy of water body extraction. The average relative error of the extraction results is 4.69%, with a determination coefficient of 0.8579. Compared with before improvement, the average relative error is reduced by 0.68%, and the determination coefficient is increased by 0.0539.