Classification of Laser Footprint Based on Random Forest in Mountainous Area Using GLAS Full-Waveform Features

Liu, Xiangfeng; Liu, Xiaodan; Wang, Zhenhua; Guo-qin, Huang; Shu, Rong

doi:10.1109/jstars.2022.3151332

Cited by 7 publications

(6 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…By combining the prior knowledge provided by echo classification with the denoising algorithm [33], the parameter selection strategy can be optimized for different scenes, which can optimize the peak signal-to-noise ratio of the target. Artificial features such as the echo width, skewness, and kurtosis, combined with machine learning methods such as support vector machines [7] and random forest models [34], can be used to distinguish between deep-water, shallow-water, and land echoes. However, hand-extracted features require specialized knowledge and extensive preprocessing.…”

Section: Learning-based Methodsmentioning

confidence: 99%

Coupling Dilated Encoder–Decoder Network for Multi-Channel Airborne LiDAR Bathymetry Full-Waveform Denoising

Zhao

Zhou

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Multi-channel airborne full-waveform LiDAR is widely used for high-precision underwater depth measurement. However, the signal quality of full-waveform data is unstable due to the influence of background light, dark current noise, and the complex transmission process. Therefore, we propose a nonlocal encoder block (NLEB) based on spatial dilated convolution to optimize the feature extraction of adjacent frames. On this basis, a coupled denoising encoder–decoder network is proposed that takes advantage of the echo correlation in deep-water and shallow-water channels. Firstly, full waveforms from different channels are stacked together to form a two-dimensional tensor and input into the proposed network. Then, NLEB is used to extract local and nonlocal features from the 2D tensor. After fusing the features of the two channels, the reconstructed denoised data can be obtained by upsampling with a fully connected layer and deconvolution layer. Based on the measured data set, we constructed a noise–noisier data set, on which several denoising algorithms were compared. The results show that the proposed method improves the stability of denoising by using the inter-channel and multi-frame data correlation.

show abstract

Section: Learning-based Methodsmentioning

confidence: 99%

Coupling Dilated Encoder–Decoder Network for Multi-Channel Airborne LiDAR Bathymetry Full-Waveform Denoising

Zhao

Zhou

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…In this study, the glacier outlines in the THR were extracted using eight methods: the band ratio method, U-Net, U-Net++, GlacierNet, SaU-Net, U-Net+cSE, LandsNet and M-LandsNet. The extraction accuracies of the eight methods were assessed using five evaluation metrics: the precision, recall, F1 score (F1), Kappa coefficient (Kappa) and overall accuracy (OA) [37]. An overview of the general workflow of this study is shown in Fig.…”

Section: Methodsmentioning

confidence: 99%

Long Time-Series Glacier Outlines in the Three-Rivers Headwater Region From 1986 to 2021 Based on Deep Learning

Chen

Zhang

Yi³

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

The deep-learning-based approach has drawn significant attention in glacier extraction due to its advantages over traditional techniques. In this study, to verify the feasibility and effectiveness of LandsNet architecture for glacier extraction, we applied a modified LandsNet (M-LandsNet) to extract the glacier outlines in the Three-Rivers Headwater Region. The band ratio method, U-Net, U-Net++, GlacierNet, SaU-Net, U-Net+cSE and LandsNet, and two scenes were used for comparison. Analysis of the two scenes indicated that the M-LandsNet had the best performance and generalization ability among the eight methods. Weather conditions had the greatest negative impact on the eight methods, followed by geographic environment and geographic location. We further extracted the glacier outlines in the Three-Rivers Headwater Region in 1986−2021 in a total of 12 periods using the M-LandsNet and through manual adjustments. The glacier area in the Three-Rivers Headwater Region has decreased by 416.40 ± 102.71 km 2 (16.53 ± 4.08%) in 1986−2021. The reduction rate (16.13 ± 5.63 km 2 a −1 ) in 2003−2021 was almost twice that (7.42 ± 5.97 km 2 a −1 ) in 1986−2003. The reduction rate of the glacier area varied among different periods and areas. Comparison with previous results indicated that the obtained glacier outline dataset in this study is reliable, and can effectively reflect the glacier area and spatio-temporal glacier changes in the Three-Rivers Headwater Region. A long time-series dataset of glacier outlines in the Three-Rivers Headwater Region in 1986−2021 is available at https://doi.org/10.5281/zenodo.5512064.This study can provide data support for the estimation of regional water resources storage.

show abstract

“…From the above content, we need to solve the initial parameters of the target response, including N, a i , µ i and σ i . By equation ( 4), if the non-Gaussian system waveform is described by one Gaussian function roughly and the echo is roughly decomposed by Gaussian components, the Gaussian components of the target response can be estimated by equation (17). Figure 2 shows the schematic diagram of the initial parameter estimation result of a synthetic aliased heavytailed signal.…”

Section: Initial Parameter Estimationmentioning

confidence: 99%

“…Based on the initial parameters, the nonlinear least squares fitting is solved by the Levenberg-Marquardt algorithm, as in equation (9), to obtain the final optimized target response, denoted as h opt (t), and the distance between the objects can be found using the position parameter µ i of the target response for ranging accuracy evaluation. Combined with equation ( 14) to characterize each Gaussian function of the system waveform, the final decomposed component f GC,i (t) can be obtained by equation (17), and the fitting accuracy of the full waveform can be evaluated. In addition, for the synthetic data, we can also evaluate the fitting accuracy of each full waveform component.…”

Section: Parameter Optimizationmentioning

confidence: 99%

“…It is of great importance to study the full waveform decomposition technique to extract more echo features. Common methods for full waveform decomposition contain deconvolution [13][14][15] and decomposition [16][17][18]. For the deconvolution method, it is generally accepted that the full waveforms can be represented as the convolution of the transmitted pulse and the target response [19].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Gaussian convolution decomposition for non-Gaussian shaped pulsed LiDAR waveform

Fang¹,

Wang²,

Zheng³

2022

Meas. Sci. Technol.

View full text Add to dashboard Cite

The full waveform decomposition technique is significant for LiDAR ranging. It is challenging to extract the parameters from non-Gaussian shaped waveforms accurately. Many parametric models (e.g. the Gaussian distribution, the lognormal distribution, the generalized normal distribution, the Burr distribution, and the skew-normal distribution) were proposed to fit sharply-peaked, heavy-tailed, and negative-tailed waveforms. However, these models can constrain the shape of the waveform components. In this article, the Gaussian convolution model is established. Firstly, a set of Gaussian functions is calculated to characterize the system waveform so that asymmetric and non-Gaussian system waveforms can be included. The convolution result of the system waveform and the target response is used as the model for fitting the overlapped echo. Then a combination method of the Richardson–Lucy deconvolution, layered iterative, and Gaussian convolution is introduced to estimate the initial parameters. The Levenberg–Marquardt algorithm is used for the optimization fitting. Through experiments on synthetic data and practical recorded coding LiDAR data, we compare the proposed method with two decomposition approaches (Gaussian decomposition and skew-normal decomposition). The experiment results revealed that the proposed method could precisely decompose the overlapped non-Gaussian heavy-tailed waveforms and provide the best ranging accuracy, component fitting accuracy, and anti-noise performance. However, the traditional Gaussian and skew-normal decomposition methods can not fit the components well, resulting in inaccurate range estimates.

show abstract

Classification of Laser Footprint Based on Random Forest in Mountainous Area Using GLAS Full-Waveform Features

Cited by 7 publications

References 52 publications

Coupling Dilated Encoder–Decoder Network for Multi-Channel Airborne LiDAR Bathymetry Full-Waveform Denoising

Coupling Dilated Encoder–Decoder Network for Multi-Channel Airborne LiDAR Bathymetry Full-Waveform Denoising

Long Time-Series Glacier Outlines in the Three-Rivers Headwater Region From 1986 to 2021 Based on Deep Learning

Gaussian convolution decomposition for non-Gaussian shaped pulsed LiDAR waveform

Contact Info

Product

Resources

About