Enhanced Geometric-Spectral Feature Learning Framework for Airborne Multispectral Point Cloud Classification

A new framework for classifying airborne multispectral point clouds (MPC) has been proposed, addressing challenges such as high-dimensional heterogeneous spatial-spectral information, unbalanced sample distribution, and inter-class spectral similarity. The framework, detailed in an arXiv paper (arXiv:2606.09123v2), introduces a two-stream feature fusion method with attention mechanisms. The first stream extracts position-encoded global spectral features using fusion self-attention, while the second stream employs multikernel point convolution and feature aggregation attention. Two MPC datasets were built for evaluation. The approach enhances representation capability of spatial-spectral features, improving land-cover classification accuracy.