A lightweight fire detection method based on improved Swin-Transformer

With the rapid development of today's society, various disasters occur frequently, among which fire causes the most serious losses, threatening human life and property safety. Traditional fire detection methods have a high false alarm rate and low reliability. In order to reduce the influence of complex environment on fire detection results and improve the accuracy of identification, this paper proposes a fire detection method based on the improved Swin Transformer. In this method, Swin Transformer is used as the basic network, and a lightweight module of CA (Coordinate Attention) attention mechanism is introduced into the backbone network, which reduces the amount of computation and improves the feature extraction ability of the model. The traditional loss function is changed to the EIOU (Efficient-IoU) loss function, which deepens the representation ability between cross-modal feature information. In order to verify the effectiveness of the improved method, it was compared with Vision Transformer and Swin Transformer. Experimental results show that the improved method has the advantages of high detection accuracy, fast detection speed and low false alarm rate.