Multimodal land use classification: Harnessing HSI and LiDAR integration

Author Identifier

Muhammad Zia Ur Rehman: https://orcid.org/0000-0001-9531-1941

Syed Mohammed Shamsul Islam: https://orcid.org/0000-0002-3200-2903

David Blake: https://orcid.org/0000-0003-3747-2960

Document Type

Conference Proceeding

Publication Title

Proceedings - 2024 25th International Conference on Digital Image Computing: Techniques and Applications, DICTA 2024

First Page

655

Last Page

661

Publisher

IEEE

School

School of Science

Funders

Edith Cowan University

Comments

Rehman, M. Z. U., Islam, S. M. S., Ulhaq, A., Janjua, N., & Blake, D. (2024, November). Multimodal land use classification: Harnessing HSI and LiDAR integration. In 2024 International Conference on Digital Image Computing: Techniques and Applications (DICTA) (pp. 655-661). IEEE. https://doi.org/10.1109/DICTA63115.2024.00099

Abstract

Recently, the integration of multiple remote sensing modalities has gained significant attention in land use classification research, offering improved performance. However, this approach comes with additional challenges such as modality-specific feature extraction and effective feature fusion. In this work, a DL-based technique is proposed that utilizes dual remote sensing modalities (HSI and LiDAR) for land use classification. The proposed technique consists of three modules: 1) a CNN-based feature extraction module, 2) Attention modules designed specifically for each modality, i.e., Convolution Block Attention Module (CBAM) and a spatial attention module for the HSI and the LiDAR features respectively. 3) A fusion module to fuse separately extracted features of both modalities. The features extracted from convolution blocks are subsequently enhanced using attention modules, later, feature-level fusion is performed, and final classification is achieved. The novel combination of these modules has demonstrated a notable performance gain over the CNN-based approaches across different classes and metrics on the Trento dataset. It achieves 98.21% average accuracy on the Trento dataset, which shows its significant potential to be applied in resource management and planning and environmental monitoring.

DOI

10.1109/DICTA63115.2024.00099

Access Rights

subscription content

Link to Full Text

Share

 
COinS