Research On Barrier Trees Identification Method Of Transmission Line Based On Hyperspectral-LiDAR Fusion Data

The society’s demand for electric energy is constantly expanding,and the geographical distribution of energy is uneven.In order to solve the problem of energy demand,a series of projects such as "west to east power transmission" have been carried out,and high-voltage engineering transmission projects have been constructed.In the no-man’s land,the lush transmission of growth,the huge growth threats that transmission lines face in these areas,when serious,will cause the grid to stop and bring huge economic losses.Tree barrier is one of the main threats to transmission lines,and the type of trees is closely related to the problem of tree barrier.Quickly obtaining the number and type of single trees has an important role in monitoring and predicting the risk of tree barrier.At present,the main inspection methods for line threats are manual inspections.Personnel from the power department enter the transmission corridor area on foot to visually remove tree obstacles.This method is inefficient and it is very inconvenient for personnel to enter mountainous areas.Hyperspectral imaging can quickly obtain the spectral characteristics of the canopy.By extracting the "fingerprint" information of the canopy,it can reflect the category from the composition of the canopy leaves.The LiDAR point cloud system can quickly obtain the spatial points of the canopy in the area.Laser point clouds can characterize the type of single wood from three-dimensional spatial structure and other aspects.This paper uses airborne hyperspectral imaging and LiDAR point cloud system to quickly collect the original hyperspectral imaging and point cloud data of a 220 k V high-voltage overhead power transmission corridor in a certain place,and conducts single tree species identification research based on fusion data.First,two types of extracted data were prepared,including filtering and denoising,repeating canopy point cloud segmentation,obtaining canopy height model(CHM),and point cloud space normalization,etc.using LiDAR point cloud data Canopy height model and normalized point cloud for single tree segmentation,and decided to fuse the hyperspectral gray graphics to correct the CHM segmentation results.Then use the segmented single tree as the smallest classification unit to extract the single tree’s hyperspectral features,including the refractive index and characteristic wavelength.The LiDAR point cloud characteristics of the single tree were extracted,including the single tree height,canopy area,canopy volume,point cloud density,point cloud intensity,and canopy topography index.Finally,a random forest and a support vector machine are selected to establish a single tree species recognition model with single features and fused features,respectively.The accurate segmentation of single trees is of great significance for the extraction of vegetation cover information.This paper uses the canopy height model(CHM)and normalized point cloud to separate the single trees.The former matches the actual number of single trees by 71.5%,while the latter The degree of matching is 73.8%.The hyperspectral grayscale image is used to make decision fusion for the CHM segmentation results,which effectively reduces the over-segmentation phenomenon and improves the matching rate by18.8%.The cloud density,intensity,and morphological index(LiDAR point cloud characteristics)have improved the recognition effect of single tree species to a certain extent,but the single tree species recognition based on a single type of feature has the highest accuracy rate of only 80%.After the fusion,the classification accuracy of the single tree category is greatly improved by the two classification models,and the accuracy of cross-validation is as high as 95.6%.Finally,a "fusion-PCA-SVM" single tree species classification model was established to identify 40 unknown single trees in the test set with an accuracy rate of 89.9%.The results show that decision fusion of hyperspectral data and LiDAR point cloud data at the stage of single tree segmentation can improve the segmentation accuracy,and feature fusion of hyperspectral data and LiDAR point cloud data at the stage of single tree species recognition can achieve single tree species.High-precision identification and fusion data can achieve accurate acquisition of vegetation coverage information,which can be used as an effective means for obtaining vegetation information and early warning of tree barriers in the transmission line area.