Research On UAV System For Forest Fire Prevention

Forest is a valuable resource in nature,research on forest fire detection techniques is signif-icant in forestry.Unmanned aerial vehicles(UAVs)are increasingly being used in forest fire monitoring and detection thanks to their high mobility.To resolve forest fire detection,the fol-lowing research work have been carried out based on UAVs technology:(1)A UAV platform for forest fire detection was constructed.According to the demand of forest fire prevention,a six-rotor UAV,equipped with hardware equipment such as SLR cameras,has been proposed to complete the task of detection of forest fires.Besides,the image processing would be performed on the computers which is equipped on UAV and ground station respectively.To resolve the problem of large data processing and high network bandwidth occupancy in the system,the network architecture of edge computing is proposed,and whole forest fire prevention system is preliminarily designed.(2)A forest fire detection algorithm,which uses the color features and wavelet analysis,is proposed based on the constructed UAV platform for application to the forest fire detection.The approach uses the color characteristics of the images taken by the UAVs and uses wavelet analysis to further process.Firstly,according to the color characteristics of forest flame and smoke,a low computational cost algorithm is adopted to extract pixels from its related regions.In order to correct the inaccuracy of color feature extraction,a two-dimensional discrete wavelet transform(DWT)is implemented to distinguish flame and the smoke area from other high-frequency noise signals.Multiple sets of experiments have proved that the algorithm proposed can effectively detect the forest flame and smoke part of the image.The good performance is anticipated to significantly improve the accuracy of forest fire detection on the basis of less computational cost and can perform real-time detection on the constructed UAV platform.(3)Based on the constructed forest fire detection UAV platform,the detection algorithms for ground-based and air-based detection based on the use of YOLOV3 is proposed.Traditional fire detection algorithms are mostly based on the RGB color model,but their speed and accuracy need further improvements.This paper proposes a forest fire detection algorithm by exploiting YOLOv3 to UAV-based aerial images.First,the forest fire image is divided into S*S units,and then according to the position of the unit,the width and height of the detection target and the coordinates of the center point are calculated by the size clustering method,and the coordinates are normalized.At the same time,the method uses logistic regression to predict the target bound-ing box score and the log loss function to predict the categories that may be included in the bounding box,calculate the confidence and then predict.The test results show that the recognition rate of the algorithm on the UAV platform is about 83%,the video detection rate can reach more than 3.2 fps,on the other hand,the recognition rate can reach 91%,and the video detection rate is more than 30fps in the ground station.This method has great advantages for real-time forest fire detection application using UAVs.(4)A preliminary design for path planning of UAVs in forest fire detection is further pro-posed.This work divides the search of forest fire prevention tasks into generalized searching and key point searching,and proposes spiral search algorithm,mower search algorithm and shortest path search algorithm based on hybrid particle swarm optimization algorithm.Simulation exper-iments with MATLAB prove that the path planning algorithm can meet the mission requirements of UAVs in forest fire prevention.