The complex working environment of coal mine drilling operations, characterized by intense illumination, water mist, and dust interference, poses significant challenges for accurate identification of drilling procedures. To enhance intelligent procedure recognition during coal mine drilling operations, this study proposes an inertial data-driven drilling procedure identification method. Firstly, the inertial response characteristics of the drill tail during various procedures were systematically analyzed. An inertial response measurement module was developed and its deployment scheme formulated. Secondly, a bidirectional long short-term memory (BiLSTM) network was employed to extract temporal features from inertial data. The classical Transformer Encoder network was improved to construct a dual-modal Transformer（DMT） feature extraction network specifically for inertial data. A BiLSTM-DMT network was designed to effectively capture features from dual-modal inertial data, enabling intelligent perception and identification of drilling procedures. Finally, field measurements were conducted underground, yielding an inertial dataset encompassing seven typical drilling procedures. The neuron configuration of the BiLSTM network was optimized through comparative analysis of multiple fusion models. Experimental results demonstrate that the proposed method achieves 98.87% recognition accuracy in training instances and 96.26% in engineering applications, significantly outperforming existing comparative algorithms. This confirms the method’s effectiveness in identifying drilling procedures with high accuracy and robustness, thereby establishing a novel technical pathway for intelligent procedure recognition in coal mine drilling operations.