A Fire Detection Method Based on Improved YOLOv5s

Aiming at the problems of low accuracy and difficulty in detecting small targets in smoke and flame detection in complex environments, an improved fire detection method based on YOLOv5s is proposed. Firstly, a new distributed shift convolution (DSConv) is introduced to reduce the number of parameters. Secondly, the Convolutional Block Attention Module(CBAM) is introduced in the YOLOv5s network model to enhance flame features and improve the efficiency and accuracy of fire detection. Finally, to address the issue of poor small object detection performance in YOLOv5, a small object detection layer is added to the feature fusion section of YOLOv5 to enhance its ability to detect small objects. The experimental results show that the improved algorithm improves the average detection accuracy by 1.4% compared to the original YOLOv5s algorithm, while also achieving higher accuracy in detecting small targets and reducing false positives and omissions.