Abstract: The treatment of black-odor water bodies is an important task in implementing the rural revitalization strategy. However, their distribution is extensive, widespread, and shows seasonal variation, leading to unclear baseline figures and periodic odors, thus becoming a difficult point for governance and supervision. In this paper, we use GF-2 as the data source and adopt the DeepLabv3+ semantic segmentation method to extract fine water body in rural areas. We analyze the spectral curve features of black-odor water bodies and normal water bodies, and build various models to identify black-odor water bodies. We conduct precision verification in Xinxiang County, Luohe City, and Xiangcheng City,and the results show that the NDBWI model has high accuracy and universality. We use this model to dynamically monitor black-odor water bodies in Xinxiang County from 2022 to 2024. The study found that the number of black-odor water bodies in 2023 is 60% less than that in the previous year, and there is a slight rebound of 6.8% in 2024, indicating that the treatment of black-odor water bodies treatment in Xinxiang County has achieved significant results in the past three years. The research results can provide effective support for precise management of rural black-odor water bodies and the promotion of rural living environment improvement.

Key words: GF-2, water extraction, rural black-odor water bodies, remote sensing identification, dynamic monitoring