Design And Implementation Of Unmanned Aerial Vehicle Forest Fire Detection System

The continuous deterioration of the natural environment is leading to the continuous increase of forest fires.Because forest fires spread rapidly,they are difficult to control once they expand.Therefore,it is particularly important to detect the fire point in time before the fire spreads.Today’s method of applying drone swarms to forest fire detection is gradually becoming common,but there are still two problems:the existing fire detection algorithms are too complex to run on the processor carried by the drone;The performance of the networking routing protocol will be degraded when it is applied to the ad hoc network of high-speed UAVs.In response to these two problems,the work done in this paper is as follows:First,the forest fire detection algorithm G-YOLOX applied to the UAV processor is proposed.For forest fire scenes,color dithering and improved Mosaic data enhancement are used to preprocess the data,and G-beck components and atrous convolution branches are introduced into the feature extraction network to reduce the model and improve the detection of small objects.precision.The experimental results show that the AP of the G-YOLOX model reaches 80.9%when performing fire detection,and the amount of parameters and calculations are almost reduced to half of the original,enabling real-time fire detection to be performed on the UAV processor.Second,a routing protocol OLSR-QER for UAV ad hoc network is proposed.Aiming at the characteristics of high-speed motion of nodes in UAV ad hoc networks,an MPR set selection algorithm that integrates link quality,device energy and node relative stability is proposed,and the format of Hello packet is modified.Simulation results show that OLSRQER outperforms traditional OLSR routing protocols in terms of packet delivery rate and average end-to-end delay when applied to UAV ad hoc networks.Third,a UAV forest fire detection system is designed and implemented.The system performs real-time detection on the video collected by the drone on the processor mounted on the drone.If it is determined that there may be a fire,it will push the video stream back to the ground for re-inspection.Otherwise,the video frame will be periodically intercepted and sent back to the ground for backup.The ground terminal realizes the management of device information and the viewing function of the returned video stream.