Abstract: The issue of eutrophication in urban landscape lakes is receiving increasing attention. Existing monitoring methods offer limited capabilities for high-frequency, large-scale water quality assessment. Taking Yuanxiang Lake in Jiading, Shanghai, as a case study, this paper proposes a lightweight image recognition method based on an improved MobileNetV2 for rapid monitoring of lake eutrophication status. A paired “image-water quality” dataset covering different seasons and eutrophication levels was constructed, including 120 UAV multispectral images and 156 in-situ water quality samples. A multi-pooling ECA attention module was introduced into MobileNetV2 to enhance the recognition and extraction of eutrophicated areas. Validation with field data shows that the model achieves an R² of 0.89 for chlorophyll-a concentration inversion, an overall eutrophication classification accuracy of 91.3%, and a transparency classification accuracy of 87.6%. Compared with conventional monitoring methods, the monitoring efficiency is improved by over 90%, and the cost is reduced by approximately 70%. This method can be deployed on relevant equipment for unmanned monitoring, providing technical support for the refined management of urban landscape lake water environments.