A lightweight attention-based Flame Detection method for Steam Boiler

To achieve real-time detection of boiler flames on embedded devices, A lightweight attention-based Flame Detection method for Steam Boiler is proposed that ensures accuracy while reducing the size of the model. This methodology not only enhances the recognition efficiency of combustion flames but also ensures model lightweighting, thereby augmenting the speed of recognition. Utilizing the Ghost module’s lower network parameters and computational complexity to enhance model efficiency, reduce model size, and expedite training and inference. Furthoremore, Utilizing the Coordinate Attention Module enhances focus on these features, thereby improving the model’s accuracy in tasks such as small-scale object detection. This ensures lightweight model design while safeguarding the precision of boiler flame detection. The precision, floating-point operations per second (FLOPs), params, and frames per second (FPS) of the improved model are $95.7 \%, 8.8 \mathrm{G}, 7.7 \mathrm{M}$, and 4.31.Compared to the original YOLOv 5 s, Although the accuracy decreased by $0.2 \%$, the params reduced by $43.4 \%$. Therefore, this detection method can be better deployed while maintaining high detection accuracy, enabling effective identification and detection of flames inside the boiler.