Abstract: Under the background of intelligent mine construction, intelligent microseismic signal processing serves as the cornerstone for precise early warning of mine dynamic disasters. The complex geological conditions and operational environments in mining areas result in microseismic signals characterized by strong noise interference, high dimensionality, and pronounced nonlinearity. Conventional automated processing methods suffer from low accuracy and heavy reliance on manual intervention for source parameter calculation, failing to meet the requirements of intelligent disaster warning. To address these challenges, this study developed a Transformer-based foundational model for intelligent microseismic processing by integrating big data analytics and deep learning theories. A comprehensive dataset containing over 300 000 microseismic waveforms was established, incorporating three key innovations: multi-scale convolutional modules for multi-dimensional feature extraction, an adaptive feature fusion strategy for noise-resistant signal representation, and a feature-aggregated multi-head attention mechanism for temporal sequence modeling. The model’s multi-task decoder simultaneously achieves intelligent event detection, P/S-wave arrival picking, and first-motion polarity determination. Experimental results demonstrate exceptional performance with 95.4% event detection accuracy, 96.6% of P-wave arrivals and 65.5% of S-wave arrivals exhibiting errors within 50 ms, and 93.38% accuracy in polarity determination, satisfying real-time processing requirements. Field application at a rockburst-prone coal face in Gansu Province confirmed the model’s engineering effectiveness, enabling fully automated processing from event detection to source localization (errors <50 ms) and mechanism inversion. This technological breakthrough establishes a robust framework for intelligent monitoring and precise early warning of mine dynamic disasters, effectively overcoming the limitations of traditional methods in complex geological environments.