| LiDAR(light detection and ranging)is a ranging device that uses laser beams to measure the distance,intensity,and other information of the laser beam’s return,enabling high-precision 3D measurements.Among them,airborne Li DAR is a type of Li DAR equipment installed on aircraft or drones,which can quickly scan the target area and obtain large-area,high-precision3 D point cloud data.It has been widely used in power line inspection.However,the geographical location of transmission corridors is special,with dense vegetation and large terrain fluctuations,which can interfere with the point cloud obtained by airborne Li DAR,resulting in noise and non-power line information in the point cloud,reducing the accuracy of extracting and evaluating the location and morphology information of transmission corridors.Therefore,it is necessary to classify and extract the point cloud data obtained by airborne Li DAR to separate the data belonging to transmission corridors,and improve the efficiency and accuracy of power line inspection.The research in this article mainly revolves around three aspects: point cloud classification of transmission corridors,classification of power tower shapes,and extraction of power lines.Based on the classification of point clouds in transmission corridors,further classification and extraction of power towers and power lines are conducted,and the accuracy of classification and extraction is improved through optimized algorithms.The main contents of this article are as follows:(1)Aiming at the problem of class imbalance in the point cloud of transmission corridors,an improved random forest classification method is proposed.Based on the characteristics of power transmission lines,point cloud features are calculated at different scales.The Relief F and sequential backward selection algorithms are used to select and weight features,retaining those with strong relevance to the classification task.Finally,the selected features are used to estimate the model’s parameters and,in combination with weighted voting principles,achieve classification of the transmission corridor point cloud.Experimental results show that compared with support vector machine,Ada Boost,and random forest algorithms,this method improves both classification accuracy and efficiency.(2)In order to extract local detailed features of power towers and achieve the classification of different shapes of power towers,an improved PointNet point cloud classification model is proposed.This model builds upon the foundation of PointNet and incorporates sampling and grouping modules,as well as a channel-wise affinity attention module.The sampling and grouping modules obtain multi-scale local neighborhood information from the point cloud and extract features from each neighborhood’s point cloud to obtain multi-scale features of the point cloud.The channel-wise affinity attention module improves attention to salient features of the point cloud.The results of the experiment demonstrate that this model can classify differentshaped power transmission towers with strong robustness and high classification accuracy.(3)Aiming at the situation where multiple power lines are arranged side by side in transmission corridors,an adaptive density clustering method is used to extract single power lines.Based on point cloud classification,elevation filtering and principal component analysis are used to roughly extract the point cloud of power lines.Then,an adaptive parameter adjustment method is used on the basis of the DBSCAN algorithm to cluster the point cloud of power lines with different densities and accurately extract single power lines.The method is validated on three different transmission lines,and the results show that the accuracy of extracting single power lines reaches 97%,and it has strong robustness against random noise. |
