Multispectral Point Cloud Classification Network Based on Multilateral Attention

Abstract

Multispectral point clouds have been increasingly applied in land cover classification. Although a variety of successful networks have been devised, they all extract local spectral-spatial features from multispectral point clouds. This paper proposes a multispectral point cloud classification network based on a multilateral attention. The network first extracts and aggregates deep local spectral-spatial features via the proposed residual multilateral aggregation module. A transformer module is then used to further learn discriminative global features. The proposed method was evaluated using two real datasets. The experimental results indicate that the proposed network performs better than some state-of-the-art classification methods.