This paper takes the drinking water source protection area of Danjiangkou Reservoir (Henan Province) as the research object. Using GF series satellite remote sensing images from spring 2022 (March to May), advanced remote sensing intelligent processing technology and deep learning algorithms are utilized to achieve accurate identification and quantitative analysis of agricultural non-point source risk sources. Through intelligent extraction of agricultural activities such as natural planting, economic forest planting, livestock and poultry breeding, and aquaculture, this study constructed a high-precision spatial distribution map of agricultural non-point source risk sources. The results show that agricultural non-point source risk sources exhibit obvious patchy distribution characteristics around the reservoir, and their distribution is closely related to land use types, terrain conditions, and human activity intensity. The research results have revealed the current situation of agricultural non-point source pollution in the Danjiangkou Reservoir Protection Area, providing a scientific basis for formulating targeted measures for water source protection and agricultural pollution control.