Objective: An improved U-Net model is proposed to automatically segment the effusion area in pericardial effusion echocardiography. Method: This article uses 70% of the publicly available dataset Pericardial Effusion experimental data collected on GitHub as the training set, 30% as the testing set, and clinical data collected from the Affiliated Hospital of Weifang Medical College as the external testing set. Firstly, the model"s generalization ability and segmentation accuracy are enhanced by introducing a multi-scale feature extraction module and Dropout2d mechanism; Secondly, the LeakyReLU activation function is applied in the downsampling process to improve the nonlinear representation of the model; Finally, reflection filling is used in the convolution layer to improve the boundary profile of the effusion area. Result:The proposed model was validated on the publicly and external testing set. Their accuracy, recall, precision, and F1 score were 96.97%, 91.47%, 69.84%, and 77.34%; 98%, 52.54%, 80.72%, and 60.86%, demonstrating its good generalization ability. Conclusion: This study provides an effective method for accurate segmentation of pericardial effusion echocardiography, which can help doctors improve the diagnosis efficiency of pericardial effusion while ensuring the accuracy.