SMWE-GFPNNet: A high-precision and robust method for forest fire smoke detection

Rui Li, Yaowen Hu, Lin Li, Renxiang Guan, Ruoli Yang, Jialei Zhan, Weiwei Cai, Yanfeng Wang, Haiwen Xu, Liujun Li

Abstract: Highlight•An attention mechanism is proposed to improve the flow of information between the swin transformer block.•A convolution method is proposed and used to eliminate redundant features, as a part of FPN.•A loss function is proposed to enhance the feature extraction effect of the smoke edge.•Forest Fire Smoke Complex Background Detection Dataset will be open-sourced with the release of the paper.