Filtering And Characteristics Of Airborne LiDAR Point Cloud On High Voltage Transmission Line Extraction Technology Research

Airborne LiDAR is an emerging surveying and mapping technology,and is another technological revolution in the field of surveying and mapping following GPS technology.At present,LiDAR technology has become a new method for the majority of scientific research and engineering technicians to solve the problem of spatial data acquisition,providing more accurate data for engineering and scientific research.LiDAR technology has made considerable progress in hardware research at home and abroad,but despite some breakthroughs in LiDAR point cloud data post-processing,there is still no very mature solution,which is still a research difficulty and hot spot.This paper is based on airborne LiDAR point cloud data of high-voltage transmission line channels.First,the point cloud data is denoised to generate basic data for filtering and feature extraction;then the mathematical morphological filtering and progressive TIN filtering are compared and analyzed in point cloud filtering Based on the processing characteristics,a combined filtering method is proposed,and the filtering experimental research and accuracy evaluation are carried out.Finally,the local features extracted in the multi-scale neighborhood and the global features extracted by Point Net are combined through a fully connected layer to perform point cloud data.Feature extraction research and analysis of results.The main conclusions drawn in this article are:(1)In places where the local terrain does not change much,the mathematical morphology moving window can be filtered more accurately,but when encountering structures with large terrain fluctuations or complex structures,the mathematical morphology moving window cannot be accurately segmented Ground points and non-ground points cause a reduction in the filtering accuracy in a large area;progressive TIN filtering is mainly affected by the selection of seed points when the initial TIN is established,and filtering errors are generated,and this error is transitive and cumulative.The point cloud data after the gross error removal is first subjected to coarse mathematical morphology filtering to segment out non-ground points with structural mutations and obvious point cloud height differences,and then the remaining point cloud data is searched to select seed points,establish a TIN,and then use The progressive triangulation filter algorithm performs multiple iteration cycles offiltering until no non-ground points are generated,and accurate ground point cloud data can be obtained.The results show that,compared with the progressive TIN filter and the mathematical morphology filter algorithm,the three types of error in the filtering strategy proposed in this paper are compared with the progressive TIN filter algorithm,except for the Data1 point cloud data with high vegetation coverage.The category error and total error were slightly higher than 1.49% and 0.3%,respectively,and the remaining error indicators were all reduced.The algorithm proposed in the study is verified,and it can show certain adaptability in complex scenes with high vegetation coverage and in flat scenes with low vegetation coverage.In scenes with simple geomorphology,the filtering error can be reduced,while in complex scenes,the accuracy is better than that of morphological filtering.(2)Point Net deep learning framework has better extraction of global features of point cloud data,but there are certain deficiencies in extracting local features of point cloud.Based on the study of the Point Net feature extraction network,this paper proposes to construct a multi-scale factor local feature extraction network while performing global feature extraction on Point Net,and then integrate the local features extracted by neighborhoods of different scales with the global features extracted by Point Net Connect,and finally output the classification score of each point in the point cloud data through the multi-layer perceptron.This improves the learning network’s learning of the local information of the point cloud data,improves the accuracy of the final feature extraction results,and has a certain generalization ability and versatility of various scenarios.The results show that on the whole,the point cloud data feature extraction of the line tower is clear,the outer contour is clear,and the feature point cloud is not mixed with other feature point clouds,all can be accurately extracted,and the transmission line cables and towers are also better The point cloud and shrub forests on the ground are classified into one category due to the small difference in point cloud characteristics;the point cloud characteristics of taller arbor forests are obviously different from the low plants,so they can be extracted according to the canopy The characteristics of vegetation.In terms of local features,the network model proposed in this paper increases the local features of multi-scale network structure extraction points,which is 6.1% higher than that of Point Net in the extraction of the main tower feature,so it can misidentify Point Net."Point cloud of Tuanbang Tower" is accurately identified as a point cloud of low vegetation.Improve the recognition rate of point cloud local features.