Abstract: In microseismic signal processing, complex noise severely interferes with the reliable identification of valid events and increases the uncertainty in first-arrival picking of P- and S-waves. To enhance microseismic signal quality and improve the robustness of first-arrival picking, this study proposes an RC-Transformer–based noise suppression method. Built upon the Transformer deep learning framework, the proposed approach leverages a self-attention mechanism to capture global signal characteristics, while incorporating residual convolutional networks to strengthen the suppression of local noise. This design enables effective attenuation of complex background noise and substantial restoration of the amplitudes of effective events. Tests on three-dimensional synthetic microseismic data, underground gas storage monitoring data, and field microseismic data from coal-roof hydraulic fracturing demonstrate that the method can efficiently suppress background noise and recover the waveform amplitudes of microseismic events. Compared with traditional algorithms, the RC-Transformer significantly improves the signal-to-noise ratio (SNR) of microseismic signals and the accuracy of first-arrival picking for P-waves and S-waves, boasts high inference efficiency, and provides an efficient and robust solution for noise suppression and effective event detection of microseisms in complex geological environments.