Research On Forest Fire Smoke Detection Based On Visible And Infrared Remote Sensing Images

Forest resources are especially critical to both the earth’s ecological environment and human survival and development.Forest fires are extremely harmful to forest resources,and their suddenness and rapid spread have always been the reason for the difficulty in handling and rescuing them.If forest fires are detected and dealt with in a timely manner at an early stage,the damage caused by fires can be greatly reduced.For this reason,researchers in various countries have started active research related to the field of forest fire prevention.Computer vision-based object detection algorithms and infrared image-based forest fire prevention research are gradually becoming hot spots.In this paper,we take the forest fire video data from the remote sensing view of UAV in Chongli District,Zhangjiakou City,Hebei Province as the research object,produce more than 12,000 smoke object detection datasets and visible and infrared image fusion datasets,and carry out the real-time target monitoring task of forest fire smoke and the research on visible and infrared image fusion in forest area,these researches help to detect the fire and suspicious person in the forest area at an early stage,and the research will help early detection of forest fires and suspicious people,and provide security for forest fire safety during the Winter Olympic Games.The main contents are shown as follows:1.A reduced information bottleneck target detection network SIR-YOLO based on the YOLOv7 algorithm is proposed.After detailing the structure of each part of the original YOLOv7,the false detection problems that appear in the original algorithm are analyzed.To this end,the SIR module is proposed with the help of information bottleneck theory,and the structure and role of the semantic information review module are described in detail.Later,the channel attention mechanism is added on this basis,and after experimental evaluation,both of them improve 7% and 8%,respectively,compared with the original YOLOv7 network in terms of AP50 metrics,having high detection accuracy and real-time detection ability.2.The end-to-end unsupervised visible and infrared image level fusion network VIF-GRL-Net is proposed as an image fusion task,where the fused images have the characteristics of high definition and semantic richness of visible images,as well as the characteristics of infrared images that are sensitive to hot spots.In the fire prevention task,it can be combined with the target detection network to obtain more accurate detection results,and also provide training data with richer semantic information for the target detection or other task.Based on the VIF-Net,this study firstly colorizes it with three channels.Then,the GRL module is added to increase the fusion effect.Finally,the MSE loss function is added to enhance the decoding process of the network.The final fused image quality exceeds that of the comparison method and has an advantage in computation time.3.Design and develop a Winter Olympic Games UAV-based intelligent forest fire prevention system in Chongli District.According to the project requirements and fieldwork data,design the system structure,explore the fire prevention system response delay problem,and give specific solutions.Later,the coordination relationship between each module of the system and the modules is introduced.The SIR-YOLO and VIF-GRL-Net proposed in this study are also applied in the system.