Linear transfer-driven adaptive asymmetric Gaussian decomposition for full-waveform LiDAR

Full-waveform LiDAR is extensively used in topographic surveying, forestry surveys, and urban monitoring, where accurate waveform decomposition is critical for high-quality point cloud generation and feature extraction. However, existing decomposition methods face two fundamental challenges: symmetric models cannot capture the asymmetric tailing characteristics of echoes, and asymmetric models with fixed shape parameters fail to adapt to diverse scattering conditions. These limitations lead to under-decomposition and over-decomposition issues that compromise decomposition accuracy in complex overlapping scenarios. To address these challenges, this paper proposes an adaptive asymmetric Gaussian decomposition (AAGD) method. First, a stable linear relationship between the broadening factor and standard deviation ratio is established through systematic convolution experiments. Based on this, an adaptive parameter adjustment mechanism is constructed to map the measured pulse characteristics to echo shape parameters for the establishment of a complete adaptive asymmetric Gaussian model. In combination with Levenberg-Marquardt (LM) optimization, AAGD dynamically adjusts decomposition parameters according to the scattering characteristics of individual echo components to ensure both mathematical precision and physical consistency of the decomposition results. Experiments on simulated and Global Ecosystem Dynamics Investigation (GEDI) data show that AAGD achieves 96.08% detection accuracy in simulated data, improving over-decomposition (0.40%) and under-decomposition (3.52%). In GEDI data, AAGD reduces root-mean-square error (RMSE) by 18.08%-41.34% compared to Gaussian decomposition (GD) and asymmetric generalized Gaussian decomposition (ASGGD), demonstrating superior performance under diverse scattering conditions.