A Swin Transformer-Based model for smoke detection

In order to minimize the damage caused by fire, it is especially important to detect smoke at the early stage of fire or when it is only produced. In this paper, we design a smoke detection network model SwinFP-Net based on the transformer algorithm, which enhances the global receptive field and reduces the arithmetic overhead through the sliding-window attention mechanism , in addition to using the feature pyramid structure to improve the detection accuracy of small targets. Validated by public dataset , the average precision of the proposed model for smoke detection is 88.5%.