Microseismic source location is the core task of microseismic monitoring. Faced with the current massive seismic data, traditional location methods are unable to meet the demand for real-time location. Therefore, this paper employs deep learning technology and proposes a microseismic source location method based on the U-Net network as the primary architecture. By integrating dual crossed attention modules and spatial dilated pyramid pooling modules, the network’s ability to extract waveform features from microseismic data is enhanced, thereby improving the prediction accuracy of the seismic source location. Finally, experiments are conducted using synthetic data generated from simple layered and complex velocity models. The prediction error accuracy of the source location is compared and analyzed with the U-Net and Att-Unet networks. The results indicate that the network model constructed in this paper surpasses other network models in terms of prediction accuracy and network performance, and also exhibits favorable prediction effects on microseismic data with low signal-to-noise ratios.