Research On UAV Remote Sensing Image Processing In Vegetation Area

In recent years,with the gradual advancement of information technology,remote sensing technology as an important direction for the development of traditional photogrammetry has made considerable progress.Different from the high cost,low efficiency,long return period of Space remote sensing and Aerial remote sensing,the UAV remote sensing effectively avoids some shortcomings of high altitude remote sensing,providing a high-efficiency,high-resolution,low-cost,low-risk resolution as one of the major ways of new low altitude remote sensing.This paper firstly introduces the remote sensing image data processing technology applied to UAV remote sensing image processing in vegetation area,including the scale invariance feature transform(SIFT)applied to the feature point extraction of remote sensing image,SIFT feature point detection processing based on GPU(Graphics Processing Unit)acceleration,Bundler algorithm of incremental three-dimensional point cloud reconstruction,three-dimensional point cloud high-density algorithm of CMVS / PMVS and so on.Based on the above technologies and the characteristics of UAV remote sensing image processing,this paper summarizes a set of different types of vegetation coverage for UAV image processing methods,mainly from the following three aspects.1,The parameters of GPU-SIFT algorithm and Gaussian difference pyramid are described in detail,and the optimal parameter setting for high-resolution UAV remote sensing image processing in vegetation coverage area is proposed.The SIFT algorithm using GPU optimization is analyzed and the performance improvement is evaluated.2,Based on the GPS information,UTM coordinates of the image points are generated and retrieved using the data structure of the K-Dimension tree to generate a pairwise perfect match based on the minimum Euclidean distance matching image,and this calculation process Using GPU parallel computing optimization.3,Dimensional reconstruction model based on SIFT feature points is proposed by using the multi-view bundle adjustment method combined with the incremental 3D reconstruction algorithm and the point cloud high-density algorithm.From the aspects of computing performance,processing speed and processing efficiency,the method of image processing for different types of vegetation area UAVs is evaluated,and the results are good.It provides a reliable way to realize real-time processing of UAV remote sensing images.