An Adaptive Filtering And Forest Canopy Height Estimation For Photon-counting Laser Altimeter Point Cloud

Spaceborne laser altimetry can obtain three-dimensional ground data and vertical structure information of ground objects.With high orbit and wide observation field,it can detect almost every corner on a global scale,and is widely used in fields such as topographic mapping,forest monitoring,and water quality monitoring.Compared to traditional threshold type and full waveform lidars,the new type of photon counting lidar uses a single photon detector with micro pulses to record all photon signals reflected by ground objects,improving the detection probability of ground objects,with higher accuracy and penetrability.It has broad prospects in the research of aerosols,clouds,vegetation,water bodies,and other aspects.However,due to its highly sensitive single photon detection mechanism,while recording the effective photon events returned by ground objects,it also records a large number of noise photons caused by equipment noise,atmospheric scattering,background interference,etc.on the laser round-trip path.Especially in forest areas,the vertical structure of ground objects is complex,the ground undulates greatly,and the probability of vegetation being measured relative to the ground is low,making it difficult to distinguish between effective photons and noise photons in photon point clouds,It affects the accurate distinction between forest canopy and ground and the estimation of forest height information,thereby limiting the scope of use of this technology.Therefore,it is necessary to study photon point cloud adaptive filtering algorithms for different terrain and surface types,and automatically extract and estimate tree height information from forest canopy and ground information.Taking ICESat-2 spaceborne laser altimetry photon point cloud data using photon counting technology as an example,this paper has carried out adaptive filtering of photon counting laser altimetry point clouds,extraction of tree canopy and ground information,and estimation of tree canopy height in forest areas.The main research is as follows:（1）An improved DBSCAN adaptive filtering method is proposed.Aiming at the photon point cloud profiles of various typical terrain objects in spaceborne lidar under different terrain,a multilevel optimized DBSCAN algorithm is used to filter noise photons,improving the accurate and efficient identification and extraction of effective photon information.For the original photon point cloud data of various typical ground objects such as mountain vegetation,water bodies,and urban areas under strong daytime background noise and weak nighttime background noise conditions,based on the DBSCAN algorithm,a Gaussian fitting coarse filtering process is added,the long and short half axes of the elliptical filtering kernel are improved,and the adaptive terrain slope fitting function is added.Comparative analysis is conducted with local distance statistical algorithm,DBSCAN algorithm,and OPTICS algorithm,respectively.Accuracy The results of evaluation on indicators such as recall rate and F1 value show that the average filtering accuracy of this algorithm for photon point cloud data under strong and weak noise conditions in mountainous vegetation,water,and urban terrain reaches 97.94%,which is the highest among the several algorithms compared in this article.Among them,the filtering accuracy of photon point cloud data in mountainous vegetation,water,and urban areas under strong noise conditions is 98.46%,98.60%,and 95.17%,respectively.This indicates that the improved DBSCAN adaptive filtering method proposed in this paper can effectively filter photon point cloud data under different complex terrain conditions（2）A photon point cloud classification method based on local distance statistics and box graph statistics optimization is proposed.For photon point clouds after noise reduction in forest areas,a two-level classification algorithm based on local distance statistics and box graph statistics is used to identify and classify ground and canopy photons,improving the extraction accuracy of forest canopy information and ground information under complex terrain.Based on the local distance statistical method,the filtered photon point cloud data in complex mountain forest regions with complex vertical structures of trees and large ground fluctuations are preliminarily classified into ground photon points and canopy photon points;Secondly,using box graph statistics,the points located above the upper edge of the box graph are removed as abnormal ground points,and compared with the results of random forest classification and histogram statistical classification.The accuracy,recall rate,F1 value,and other indicators are used for evaluation.The experimental results show that the classification accuracy of the proposed method is98.7%and 97.4%in dense and sparse vegetation scenarios,respectively,higher than 95.3% and 93.7%of the random forest algorithm and 84.6%and 90.5%of the histogram statistical classification algorithm.This indicates that the classification algorithm in this paper can effectively classify ground photon points and vegetation photon points in both dense and sparse vegetation scenes.（3）For forest canopy height estimation under complex terrain,overlapping moving window method is used to identify canopy top photons from the classified vegetation canopy photons,and estimate the height of the canopy top,thereby estimating the forest canopy height within the region.For classified vegetation photon point cloud data in a forest area with complex terrain,the overlapping moving window method combined with the upper quartile in box graph statistics is used to identify the photons at the top of the canopy from the vegetation canopy photons.The photon point cloud data obtained from the ICESat-2satellite in Hainan Island from 2019 to 2022 under different vegetation coverage is selected for testing,and specific percentile heights(H₂₅,H₅₀,H₇₅,H₉₀,H₁₀₀)of the vegetation canopy are estimated,The correlation coefficient and root mean square error indicators are selected for error analysis.Experimental results show that the proposed method can effectively estimate forest canopy height under complex terrain.