To address the issues of large parameter volume,low efficiency,and high sample requirements in current weather radar networks for biological echo extraction,this paper proposes an improved biological echo extraction network:MFF-PSPnet(Multi-Feature Fusion Pyramid Scene Parsing Network).The network introduces attention mechanism modules and edge extraction modules into the MobilenetV2 backbone,and then fuses all features to enhance the network's segmentation capability in terms of both main body extraction and edge detail characterization of biological echoes.The proposed MFF-PSPnet is a lightweight network with a 89.78% reduction in parameter volume compared to typical biological echo extraction networks.Furthermore,it requires fewer samples than other models,making it more suitable for small sample environments.Ablation and comparative experiments on historical data demonstrate that MFF-PSPnet achieves an accuracy of 98.1% and an Intersection over Union(IoU) of 94.8% in extracting biological echoes.The improved model effectively extracts biological echoes from historical data of the new generation of Chinese weather radars and better adapts to small sample environments,making it applicable to mobile devices.