Improved Smoke and Fire Detection Based on YoLoV8

The early stages of a fire involve complex and variable smoke and flame patterns, with flames often being small in size. Existing object detection algorithms typically face challenges such as slow detection speeds and low accuracy. This paper proposes an improved fire object detection system based on YOLOv8, which incorporates the CBAM attention mechanism to enhance multi-scale features and stabilize weights, and introduces various data augmentation methods. Additionally, to better handle different object sizes, the previous loss function has been optimized. The improved YOLOv8 model demonstrates increased detection accuracy, with experimental results showing a 3% improvement in mAP.