Research On Point Cloud Filtering Algorithm And Single Tree Information Extractio

In recent years,the theory and technology of lidar have developed rapidly,and the methods of acquiring point cloud data have become more and more abundant,and have been widely used in surveying and mapping,forestry,planning,electric power and other industries.Airborne Light Detection and Ranging(ALiDAR)is widely used in forest resource exploration due to its wide measurement range,high efficiency,and strong ability to penetrate the forest canopy.The data processing of airborne lidar is an important part of the ALiDAR workflow,and it is the basis of point cloud filtering to obtain a Digital Elevation Model(DEM)and subsequent data analysis.When using ALiDAR technology for forestry resource exploration in sloping forest areas,how to make the data processing algorithm more perfect and how to better reflect the terrain features while ensuring the accuracy are two major difficulties.This research focuses on point cloud filtering algorithm and single tree segmentation algorithm in hillside forest area.Firstly,the existing filtering algorithms are compared and studied,and then the optimal combination of filtering methods based on elevation normalization is proposed.Finally,according to the influence of the filtering results on the accuracy of single tree segmentation,the experimental analysis is carried out,which verifies the effectiveness of the proposed method in this paper.The main contents and achievements of this paper are as follows:(1)Comparison and research on existing filtering algorithmsResearch on the adaptability of existing filtering algorithms to terrain,comprehensively analyze the advantages and disadvantages of each algorithm,and optimize the shortcomings of the algorithm.This paper first summarizes the existing filtering accuracy evaluation methods,and selects the Kappa coefficient value as the evaluation of the accuracy of each filtering algorithm.At the same time,12 research areas with complex terrain were selected and divided into three groups according to different features and terrain features: vegetation areas with large slopes,relatively flat and densely populated urban areas,and terrain discontinuous areas with a certain slope.Experiments show that the average Kappa coefficient of the five filtering algorithms is over 65%,of which the multi-level moving surface filtering algorithm is the highest at 85.51%,and the cloth simulation filtering algorithm is second at78.13%,the above two algorithms are not effective in some areas with large slopes and discontinuous terrain,and cannot take into account the filtering accuracy and the preservation of terrain features at the same time.(2)Propose an optimized combination of filtering methods based on elevation normalizationThrough the research on the filtering algorithm,it is found that the filtering accuracy of the area with large slope is generally poor.Therefore,combining the advantages of Multi-level Moving Surface Filtering(MMSF)and Cloth Simulation Filtering(CSF)algorithms,a filtering method based on elevation normalization is proposed.This method does not directly use the two algorithms to filter,but first sets the index window based on the slope level for the point cloud data;then obtains the approximate DEM through MMSF filtering,and then eliminates the influence of terrain based on the normalization of ground points;Finally,CSF filtering is used to obtain more accurate classification results.This method has a very significant filtering effect on areas with large slopes.Using the normalization-based filtering algorithm to filter the above three groups of experimental areas,the I error,II error and total error obtained by using the MMSF and CSF algorithms alone are reduced by 3.26%,7.60%,5.95% and 2.32%,17.29%,7.75%.In particular,the filtering effect is significantly improved in forest areas with slopes greater than 20° and discontinuous areas between15° and 20°.It is verified by comparison that the proposed method has the best filtering accuracy while retaining topographical features.(3)Study the effect of classification results of different filtering algorithms on the accuracy of single tree segmentation and information extractionThe influence of the classification results of different filtering methods on the segmentation accuracy of single tree in the forest area with large slope is mainly studied.The single tree segmentation algorithm based on point cloud and the single tree segmentation algorithm based on Canopy Height Model(CHM)are described,the relationship between the filtering results and the accuracy of single-tree segmentation was verified.Under the same parameters,the classification results of MMSF,CSF and normalized filtering algorithms were directly used in single-tree segmentation experiments,experiments show that the classification results of the filtering algorithm based on elevation normalization have obtained high accuracy in the single-tree segmentation experiment: Rprec is 81.00%,and F＿Score is 0.85.The optimal segmentation result is extracted for the height of a single tree.Since the segmentation result of a single tree does not completely match the position of the measured trees,in this paper,it is proposed to replace the extraction of the height of a single tree with the average stand height of the extracted sample plot.The correlation coefficients between the measured average height of the tree stand and the estimated average height of the tree stand in the fitted sample plot reach 0.7543 and 0.8702.And the estimated average heights of the sample plots are 26.88 m and 25.36 m,which are closer to the measured average tree heights of 27.38 m and 25.42 m.This shows that the method extracts trees with high precision and the results of single-tree segmentation are reliable.