Abstract: Abstract: In order to improve the accuracy and speed of forest fire detection and enhance the practicability of forest fire detection model, an improved YOLOv5 forest fire detection algorithm was proposed. In this algorithm, the improved MobileViT was used as the backbone network of YOLOv5, so that the network could extract forest fire feature information more effectively. Meanwhile, in order to further reduce the complexity of the model, depthwise separable convolution was used to replace the common convolution in the model, and Mosaic data enhancement method was introduced in the training stage to improve the generalization of the model. The results show that the forest fire detection accuracy of the improved model is increased by 2.25%, mAP by 4.48%, and detection speed by 4 frames/s. Both the detection accuracy and detection speed have achieved good results. The improved model can detect forest fire well and improve the practicability of forest fire detection model. The algorithm in this paper is more competent for the task of forest fire detection.

Key words: forest fire detection, YOLOv5, MobileViT, depthwise separable convolution