Research On Campus Forest Fire Real-time Monitoring System Based On Deep Learning

In recent years,forest fires have occurred frequently around the world,causing serious economic losses and casualties.Traditional forest fire detection methods have problems such as low accuracy and high cost.Therefore,how to accurately and efficiently detect forest fires is currently a research focus in the field of forest fire detection.With the development of computer technology,deep learning based on object detection technology has been applied to forest fire detection tasks by many researchers due to its advantages of high detection accuracy and fast detection speed.In response to the current problems of low recognition accuracy and high computational complexity in the forest fire detection algorithms,this paper proposed an improved G-YOLOv5n-CB forest fire detection algorithm based on YOLOv5 algorithm,starting from improving algorithm recognition accuracy and reducing computational complexity.Based on this algorithm and robot technology,a real-time fire monitoring system for campus forest land was designed.This system captured image information through a camera while using an unmanned vehicle to automatically navigate.The algorithm proposed in the paper was then deployed on the Jetson Nano hardware platform carried by the unmanned vehicle to achieve real-time detection of forest fires.Finally,the detection results ware transmitted in real-time to the upper computer interface of the system for display.The main research work and innovative points of the paper are as follows:(1)A YOLOv5n-CB forest fire detection algorithm based on improved YOLOv5 n was proposed to address the issue of poor recognition accuracy in forest fire detection algorithms.Based on the YOLOv5 n model,a CBAM attention mechanism was introduced at the connection between the backbone network and the neck network,allowing the network to focus on effective information,achieving the goal of introducing small parameter quantities to effectively improving network performance.Secondly,the weighted bidirectional feature pyramid network(Bi FPN)was introduced to improve the neck network,strengthen context information fusion,and enhance the feature extraction ability of the target.Experiments shown that the proposed YOLOv5n-CB algorithm had increased the m AP value index by 1.4% compared to the original algorithm on the self-made forest fire dataset,indicating a significant improvement in detection performance.(2)A G-YOLOv5n-CB forest fire detection algorithm based on a lightweight framework was proposed to address the issue of poor real-time performance caused by excessive computational complexity and model volume in forest fire detection algorithms.Lightweight improvements were made to the feature extraction network of the original algorithm using Mobile Net V3 network and Ghost Net network respectively.Experiments shown that the lightweight strategy combined with the Ghost Net network significantly reduced the number of parameters and computation while maintaining high detection accuracy,resulting in better model lightweight performance.Comparative experiments were conducted between the improved model and other deep models such as Faster R-CNN.The results showed that the improved model significantly reduced model volume while maintaining high recognition accuracy,verifying the effectiveness of the improved algorithm.The improved G-YOLOv5n-CB model was deployed on the Jetson Nano platform for testing,with a detection speed of 15 FPS,which basically meets the real-time requirements of forest fire detection.(3)Designing a real-time monitoring system for campus forest fires.The system deployed the forest fire detection algorithm proposed in the paper on the Jetson Nano hardware platform carried by unmanned vehicles,achieving real-time detection of forest fires.The system used the Cartographer algorithm to construct unmanned vehicle navigation maps,using A * algorithm and DWA algorithm under the Move＿base navigation function package respectively perform global and local path planning for unmanned vehicles,achieving the navigation function of unmanned vehicle target points in the campus forest scenarios.Based on Py Qt5,a system upper computer interface had been developed to achieve real-time display of forest fire detection results and action control of unmanned vehicles.The comprehensive testing results of forest fires and automatic navigation modules indicated that the campus forest fire real-time monitoring system designed in the paper can accurately and real-time detect forest fires in campus forest scenarios and had practical application value.