Tree Species Classification Of Power Line Corridor Based On Airborne LiDAR And Aerial Images

The growth of trees in the transmission corridor will pose a threat to the power line.Acquisition of tree species information of the transmission corridor has an important role in the intelligent warning system for hidden dangers of tree barriers.This study considered the advantages of airborne Li DAR point cloud data and high-resolution aerial imagery.Based on the object-oriented principle and extracting multiple types of features,with nonparametric machine learning algorithms as the core to classify single tree species.This study selected the two power transmission corridor areas in northeast Chizhou,Anhui Province as the research area.Firstly,the Li DAR data is processed to obtain the Canopy Height Model(CHM)and optimized it;In view of the current background influence and the problem of over-segmentation in the canopy extraction,used the visible-band difference vegetation index(VDVI)and bilateral filtering to extract and optimize the canopy area for aerial orthophotos,then used a multi-scale segmentation method to extract the single tree canopy with combining the CHM and the orthophotos containing only the canopy,Then took the single tree canopy as the object to extract the spectrum,texture,shape and height features,and used the XGBoost algorithm for feature importance ranking and feature selection;Finally,using three non-parametric classifiers that including Random Forest(RF),Support Vector Machine(SVM)and Artificial Neural Network(ANN),designed 12 classification schemes for single tree species classification and accuracy evaluation,analyzed and compared the impact of multi-source data combination and feature selection on tree species classification accuracy,and evaluated the ability of different classifiers to classify existing tree species.The results show that the use of VDVI and bilateral filtering can reduce the impact of background features and the excessive segmentation phenomenon caused by the significant texture features within a single canopy,and the accuracy of the obtained canopy segmentation results is above 80%;using the ESP2 scale evaluation tool to determine the optimal segmentation scale parameter can improve efficiency and reduce the error of manually determining the parameter.The schemes with the highest classification accuracy in the two research areas are the same.Both are based on the combination of Li DAR data and aerial orthophotos,then used the ANN classifier after feature selection.The overall accuracy is 86.19% and 81.37 respectively.