End-to-End Full-Waveform Echo Decomposition Based on Self-Attention Classification and U-Net Decomposition

Abstract

Different from conventional decomposition methods which utilize several steps to obtain the final result, a selfattention based neural network, Attention Full-waveform Decomposition Network (AFD-Net), is discussed in this paper for endto-end full-waveform LiDAR signal decomposition. In existing LiDAR waveform decomposition methods, complicate functional models are used to fit echo components. Thus, the echo decomposition problem can be translated into a function approximation task. Recent studies present great progress in estimating the parameters of fitting models, hence in the final decomposition results. However, the shape of received echo components are always irregular. None of the parametric functional models can fit the received echo components perfectly, which leads to unavoidable errors in the initial step of echo decomposition. In this paper, we propose an end-to-end net work AFD-Net to solve the echo decomposition problem without assuming any parametric functional models. AFD-Net is consisted with two modules, the classification module and the decomposition module. The former module is used to determine the number of echo components in a received waveform. Then the decomposition module is used to output the echo components. By experiments, we have a classification accuracy 96% using the first module. The average R 2 coefficient for the decomposed echo components is 0.9799. In addition, there are no public datasets for the waveform decomposition task available. Thus, another contribution of our work is to develop a tool to generate synthetic full-waveform LiDAR signals, which can help researchers to construct their own dataset for related works. All of our source codes are available in the following site: https://github.com/ZedFm/AFD-Net

Progressive Gaussian Decomposition of Airborne Bathymetric LiDAR Waveform for Improving Seafloor Point Extraction

“…Accordingly, suitable parameterized functions for each returned waveform should be assumed to distinguish the meaningful components accurately. A Gaussian model is generally utilized for waveform decomposition [9,16]. The approximated parameters (e.g., amplitude, center, and width) of each decomposed Gaussian component are related to the target's physical properties and can be used as waveform features for further applications such as point classification [17][18][19] and land-water discrimination [20,21].…”