Research On Early Forest Fire Detection Method Based On UAV Vision

With the advancement of science and technology,unmanned aerial vehicle(UAV)is no longer limited to military uses.New explorations in the civilian field have been begun in recent years.The drones equipped with cameras have started been investigated more for monitoring forest fires.Therefore,the research of video fire detection technology is of great significance.Its characteristics are non-contact,rapid response,large detection range,and active visualization.This paper uses the forest fire detection based on UAV vision as the research object and the following research works have been carried out:(1)The smoke detection based on UAV vision is introduced.The UAV equipped with a CCD camera is used to capture video images.The acquired images are enhanced to make the smoke more distinctive and facilitate the subsequent detection work.The next section deals with smoke's saliency detection algorithm,the low rank matrix recovery algorithm,including the principal component analysis algorithm,and the principles of Mean Shift clustering algorithm and block matching algorithm.Through the image detection and segmentation results in several different scenarios,low-rank matrix recovery combined with Mean Shift clustering and block matching algorithm can well detect smoke.Finally,feature extraction is performed on the segmented target area,and the extracted features are compared.(2)Based on the forest fire flame detection of drone vision,the flame saliency detection algorithm and the analysis of different color models are introduced.Then the saliency detection results of different scenes in different color models are given.From experimental results,the detection results under the Lab color model are good,so the flame image is segmented in the Lab color space,and a good segmentation result is obtained.Finally,feature extraction is performed on the segmented flames,and the features under different scenes are compared and analyzed.It can be seen that the features selected in this thesis can well distinguish flames and flame-like objects.(3)The application of support vector machine(SVM)algorithm in forest fire identification is introduced.Firstly,the principle of SVM and the kernel function are briefly introduced,and the parameters that need to be changed are explained in detail.Next,using the forest fire smoke and flame characteristics extracted in Chapters 2 and 3,the SVM training and testing are conducted,and the classification results of the support vector machine have been obtained.According to the final classification results,the detection algorithm and the characteristics of forest fire used in this thesis can effectively detect and identify forest fires,providing a basis for the construction of forest fire monitoring systems.