YOLOv8-MFT: A Fire Detection Algorithm Based On Improved YOLOv8

Fire detection is a key topic in the field of public safety. Traditional sensor methods are susceptible to interference in complex environments, resulting in high false alarm rate and lack of real-time. Existing techniques are susceptible to background interference, resulting in false positives, false negatives and low detection accuracy. In order to solve these problems, this paper proposes a fire detection algorithm based on improved YOLOv8. Improved model YOLOv8 - MFT focused on multi-scale feature extraction and mission alignment, including the main CSP-MSEIE module (Mutil Scale Edge Information Enhance) in alternative c2f module, to strengthen the edge of the backbone network awareness, The FDPN (Focusing Diffusion Pyramid Network) is used to improve the efficiency of context information fusion, and the TDADH Detection Head (Task Dynamic Align Detection Head) is designed to reduce parameter redundancy. The improved algorithm is significantly better than the baseline model in detection accuracy and robustness. The accuracy, recall rate, mAP50 and MAP50-95 of the improved algorithm (YOLOv8-MPT) and the original model yolov8n are 5.4%, 2.4%, 4% and 2.9% higher than the original YOLOv8n, and the number of parameters is reduced by about 18%.