Abstract: The traditional image classification methods mostly use endmember spectrum to classify ground objects, however, the spatial structure information of images is ignored. In this study, an improved spectral angle mapper image classification method combining hybrid pixel decomposition means such as endmember type selection and pixel purification and watershed image segmentation algorithm is proposed, and the mangrove ecosystem is classified based on spectral feature analysis and ground survey using GF-1 remote sensing image data with Shankou Mangrove National Nature Reserve as the study area, and the classification accuracy is also analyzed. The research results show that the improved spectral angle mapper classification method is effective in classifying GF-1 images, taking into account the specificity of the complex spectral composition of the ground class and effectively avoiding the fragmentation of the results, and the overall accuracy reaches 95% with the KAPPA coefficient of 0.944, which proves its application potential in mangrove remote sensing image classification and information extraction, and lays a foundation for the operational remote sensing monitoring of mangrove ecosystem.