Filtering Photon Cloud Data in Forested Areas Based on Elliptical Distance Parameters and Machine Learning Approach

Abstract

The Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) was successfully launched. Due to its small spot size, multibeam configuration, high sampling rate, and strong immunity to terrain slopes, it has been regarded as a powerful tool for forest resources surveying and managing. However, the ICESat-2 photon cloud data contain considerable background photons, which discretely distribute in the background space of signal photons. Therefore, it is necessary to filter these noise photons. In this study, photons are divided into three categories: signal photons, noise photons far away from signal photons, and noise photons adjacent to signal photons. Based on the existing research, forward and backward elliptical distances were used to express the spatial relationship between two photons, and backward local density (BLD) was used to describe the density distribution of the photons. However, the single statistical parameter cannot clearly distinguish three types of photon cloud. Therefore, forward local density (FLD) and neighboring forward local density difference (NFLDD) also were defined to describe the density distribution of the photons. Finally, by combining the support vector machine (SVM), the above three density parameters were used to classify the photons by signal and noise photons. The proposed method was validated with photon cloud data acquired by the Simulated Advanced Terrain Laser Altimeter System (MATLAS), the Multiple Altimeter Beam Experimental Lidar (MABEL), and the ICESat-2 systems over different forested areas. The results demonstrated that the proposed method can well remove the noise photons and retain the signal photons without depending on any statistical assumptions or thresholds. The comprehensive accuracy of the three test sites was 0.99, 0.98, and 0.99, respectively, which was higher than those of the existing method. In addition, the total errors corresponding to the three test sites were about 0.4%, 0.5%, and 1.0% respectively, which were lower than those of the existing method.

Research on Multilevel Filtering Algorithm Used for Denoising Strong and Weak Beams of Daytime Photon Cloud Data with High Background Noise

“…For example, Meng et al [33] proposed a back propagation neural network denoising algorithm suitable for ICESat-2 photon data of shallow island and reef areas, and the results showed that the denoising algorithm has good denoising results in shallow island and reef areas. Li et al [34] used the support vector machine algorithm to classify signal photons and noise photons based on extracting elliptical distance parameters from ICESat-2 photon data in forest areas. The results showed that this method can effectively remove noise photons without relying on any statistical assumptions or thresholds, and the accuracy is higher than 0.98, which is superior to existing algorithms.…”

Estimation of Forest Canopy Cover by Combining ICESat-2/ATLAS Data and Geostatistical Method/Co-Kriging

“…ICESat-2 ATLAS photon-counting LiDAR emits weak signal pulses, its data contain a large number of background noise photons caused by solar background [42]. For the accuracy of the ensuing inversion of the forest parameters and mapping in this situation, the elimination of these noise photons is crucial [43]. For effectively controlling noise in photon-counting LiDAR data, a number of noise filtering algorithms have so far been established by earlier studies.…”

A Density-Based Multilevel Terrain-Adaptive Noise Removal Method for ICESat-2 Photon-Counting Data

“…Li et al [ 28 ] and He et al [ 29 ] proposed a relative neighboring relationship (RNR) and local outlier factor algorithm with a rotating search area (LOFR) to improve the local parameter statistics algorithm. For supervised machine learning techniques, Li et al [ 30 ] and Chen et al [ 31 ] proposed the neighboring forward local density difference (NFLDD) and K-nearest neighbors distance to characterize the differences between the signals and noise. Supervised methods, such as random forest and support vector machine, were used to classify the photons, which achieved good accuracy, but these methods rely on a large number of training samples.…”

“…There are no unified tools and standards for preparing samples, making it difficult apply them widely. However, there is residual noise after single-level filtering algorithms [ 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 ], especially noise adjacent to the signal in complex terrains.…”

“…In this study, we propose a density-based multilevel terrain-adaptive noise removal method (MTANR) with a coarse-to-fine denoising strategy to address the issue. It is challenging to remove the noise adjacent to the signal and the removal requires a nearly noise-free environment; therefore, we categorized noise into three types [ 25 , 30 ]: cloud-caused high-altitude distant noise, sparse noise in medium distance, and edge noise adjacent to signal, as shown in Figure 1 . Based on the classification of noise photons, the denoising process was divided into three steps: (1) Distant noise and part of sparse noise were removed using a histogram-based successive method which gradually tightened conditions as a coarse denoising process.…”