Abstract: Coal–rock identification technology plays a crucial role in enabling adaptive cutting for tunneling equipment. It serves as a fundamental support for enhancing both the efficiency and intelligence of modern coal mining operations. To overcome the limitations of existing identification methods—which often depend on ideal laboratory conditions and lack engineering adaptability—a coal–rock recognition model based on multi-source data fusion was developed, specifically designed for the complex and variable conditions of underground excavation. A corresponding hardware–software system was also developed for integration with bolter miner equipment, facilitating the practical deployment of this technology in real-world scenarios. To ensure real-time signal processing in the field, we applied a wavelet denoising algorithm to vibration signals generated during cutting, extracting multidimensional feature parameters. These were integrated with additional sensor data, including cutting current, groove feed rate, and drum height, to construct a comprehensive feature library for coal–rock identification. Secondly, the hyperparameters of the BP neural network are optimized using the improved grey wolf optimizer (IGWO) algorithm, resulting in a coal–rock identification model based on fused multi-source features. The model achieved a classification accuracy of 97.96% on the validation dataset, demonstrating excellent performance and generalization ability. To meet the demands of on-site deployment, we designed an edge computing software platform using LabVIEW and MATLAB. This platform supports real-time data acquisition and storage, signal processing, coal–rock recognition, and data transmission. Additionally, a Unity-based visualization system was developed for the upper-level interface, enabling real-time display and interaction with edge data, thereby improving system operability and visual clarity. Finally, industrial-scale tests were conducted in an actual tunneling workface to validate field deployment. Experimental results demonstrate that the proposed coal–rock identification system achieves high recognition accuracy and robust engineering adaptability under real-world operating conditions. The height recognition error of coal–rock boundaries on both sides of the roadway remains within 5 cm, validating the accuracy and practical utility of the method. This system provides critical technological support for intelligent cutting and floor-parallel excavation using bolter miner equipment, and offers reliable perceptual and decision-making capabilities essential for safe and efficient operation of intelligent tunneling machinery in modern coal mines.