MMS-YOLO: a Multi-scale Enhanced and Adaptively Optimized Method for Precise Tunnel Fire Detection

To address the problems of low accuracy and insufficient video image data in traditional fire detection methods, a tunnel fire detection model MMS-YOLO based on improved YOLOv8 is proposed. In the backbone network of YOLOv8, the model inserts Multi-scale Dilated Convolution (MDC) into the deep feature layers to improve the detection performance for early fires and complex scenarios; introduces a Median-enhanced Spatial and Channel Attention Block (MECS) to enhance global information capture and noise robustness; and adopts a Spatial Adaptive Feature Modulation Network (SAFMN) to optimize feature expression and context modeling in multi-scale target detection. Compared with traditional models, this model increases the mean average precision (mAP50-95) by 2.2%, precision by 3.39%, and recall by 4.85%, effectively improving the accuracy and robustness of tunnel fire detection and possessing good engineering application potential.