Research On Image Recognition Method Of Early Stage Of Forest Fire Based On Deep Learning

Harm of forest fires to human and ecological resources is very serious,and the timely detection of fires in the early stage of forest fires has attracted great attention from all countries.In the early stage of fire,due to insufficient combustion of combustion,a smaller flame and a large amount of smoke will be produced,and in the complex forest environment,the smaller flame is easily partially or completely blocked,and the fire cannot be predicted earlier,which requires a large economic cost.Based on the above background,this paper proposes an improved YOLOv5 algorithm,which is improved in the YOLOv5 Backbone network and Neck network respectively,so that it has better accuracy in image recognition in the early stage of forest fire.Insert CBAM into the C3 module in the YOLOv5 backbone network and add BotNET at the tail of the backbone network to enhance the detection capability of YOLOv5 for the initial stage of forest fires.The small target detection layer is added to enhance the detection ability of small targets in the early stage of forest fires,and because the channel length of the feature fusion network increases after the small target layer is increased,the information is lost,so the Bi FPN feature fusion network is integrated again to strengthen the image recognition ability of the initial forest fire.Finally,control test was conducted between the modified YOLOv5 and the original YOLOv5,and the experimental results show that the average detection accuracy of the improved YOLOv5 is increased by 3.8% when the fire target is small and the characteristics of the fire target are not obvious,which can prove that the improved YOLOv5 method proposed in this paper has a certain effect on image recognition in the early stage of forest fire.