Analysis and comparison of denoising methods for photon-counting laser data

Abstract

The Ice, Cloud, and Land Elevation Satellite-2(ICESat-2) is equipped with the advanced topography laser altimeter system (ATLAS). Product data has background noise due to atmospheric scattering, solar radiation, instrument noise, etc. So, data denoising is necessary before processing and applying the product data. This paper experimented with ICESat-2 land data using three methods: the K-nearest neighbor distance algorithm, DBSCAN algorithm, and DRAGANN algorithm. The experimental results show that the DRAGANN algorithm has the best accuracy. The DBSCAN algorithm is suitable for dealing with land areas with much vegetation, and the KNN algorithm is suitable for dealing with land areas with less vegetation. Different surface types greatly affect the accuracy of these methods for denoising.