250981 Enhanced Publishing |LightMamba: A Lightweight Mamba Network for the Joint Classification of HSI and LiDAR Data

Serious declaration: If this dataset is used in papers, books, academic reports, or other works, please cite the following data paper references: Liao Diling, Lai Tao, Huang Haifeng, Wang Qingsong LightMamba: A lightweight Mamba for joint classification network of HSI and LiDAR data [J]. Journal of Electronics and Information Technology, pre published doi: 10.11999/JEIT250981 LIAO Diling, LAI Tai, HUANG Haifeng, WANG Qingsong. LightMamba: A Lightweight Mamba Network for the Joint Classification of HSI and LiDAR Data [J]. Journal of Electronics & Information Technology, in press. doi:10.11999/JEIT250981 Authors: Liao Diling, Lai Tao, Huang Haifeng, Wang Qingsong Unit: School of Electronic and Communication Engineering, Sun Yat sen University DOI：10.11999/JEIT250981 Original text: https://jeit.ac.cn/cn/article/doi/10.11999/JEIT250981Corresponding author: Liao Diling, [e-mail] Open source date: December 30, 2025 Fund Project: National Natural Science Foundation of China (62273365), "Xiaomi Young Scholars"Project Open source content LightMamba: A Lightweight Mamba for HSI and LiDAR Data Joint Classification Network - Source Code Abstract: The joint classification of hyperspectral image (HSI) and light detection and ranging (LiDAR) data is a key task in the field of remote sensing. It significantly improves the accuracy of ground object recognition by fusing rich spectral information and accurate three-dimensional structural information. However, existing joint classification methods based on deep learning (DL) are still limited by high model computational complexity. Therefore, this article proposes a novel lightweight Mamba network. The core of this network is the introduction of an advanced State Selective Model (SSM), whose linear computational complexity enables efficient modeling of long-range contextual dependencies in remote sensing data. Firstly, the multi-source alignment module is used for feature extraction and spatial spectral alignment of heterogeneous HSI and LiDAR data to provide consistent feature representations; Secondly, the multi-source lightweight Mamba module uses LiDAR elevation information as a guide and adopts lightweight design to fuse dual stream sequences, efficiently modeling long-distance dependencies; Finally, we also designed a classifier based on MLP and output the classification results. The experimental results on multiple publicly available benchmark datasets show that compared with current advanced methods, LightMamba has achieved significant improvements in classification accuracy while maintaining lower computational complexity. This demonstrates the enormous potential of the Mamba based architecture in remote sensing multi-source data fusion and classification tasks.