By using GF-3 synthetic aperture radar images, the physical mechanisms and statistical characteristics of dual-polarization backscatter signals across various land cover types(water bodies, mountains, buildings, etc.) were analyzed. A water area extraction method based on the minimum backscatter coefficient was proposed, and the auxiliary optimization effects of DEM and image texture features were investigated. During the No. 2 floods of the Beijiang River in 2022 and 2024, the inundation extents of the two floods were successfully extracted using the method through comparison and verification with optical images. Results indicate that the riverside farmland and villages of Dazhan Town in Yingde City at the confluence of the Beijiang and Wengjiang rivers, Qingqiao Town in Qingcheng section of the Pajiang River, and Longtang Town in Qingcheng section of the Dayan River were severely affected by floods. The flood inundation extent in 2022 was larger. Specifically, Dazhan Town in Yingde section of the Beijiang River and Qingqiao Town in Qingcheng section of the Pajiang River suffered severe damage in 2022, while Dazhan Town in Yingde section of the Beijiang River and Zhangbei Town in Yingde section of the Lianjing River were significantly impacted in 2024.