Related Experiment Video
Updated: Jul 8, 2026
07:34
Excitation-Scanning Hyperspectral Imaging Microscopy to Efficiently Discriminate Fluorescence Signals
Published on: August 22, 2019
Spectral-Spatial Dynamic Scan Mamba for Multi-Source Remote Sensing Data Classification
Summary
This study introduces a novel Spectral-Spatial Dynamic Scan Mamba (SDSM) for advanced multi-source remote sensing data classification. The SDSM method improves spectral-spatial feature extraction and cross-modal fusion for enhanced ground object categorization.
Area of Science:
- Remote Sensing and Geospatial Analysis
- Artificial Intelligence and Machine Learning
- Data Fusion and Classification
Background:
- Multi-source remote sensing data classification integrates diverse data (hyperspectral image, LiDAR, SAR) to categorize ground objects.
- Existing Mamba-based methods use fixed scanning patterns, limiting spectral-spatial information characterization.
- Current fusion techniques often overlook complementary inter-modal characteristics, using simple concatenation or attention.
Purpose of the Study:
- To propose a Spectral-Spatial Dynamic Scan Mamba (SDSM) for improved multi-source remote sensing data classification.
- To address limitations in spectral-spatial feature extraction and heterogeneous feature fusion.
- To enhance the characterization of complementary information across different remote sensing data modalities.
Main Methods:
- Developed a dynamic scan Mamba network for adaptive spectral-spatial feature extraction from multi-source data.
- Introduced a dynamic scan module to dynamically capture salient spatial and spectral information.
- Proposed a bidirectional cross-modal fusion rule incorporating a global-local frequency feature extraction module for guided heterogeneous feature fusion.
Main Results:
- The proposed SDSM method demonstrated superior performance on four benchmark multi-source remote sensing datasets (MUUFL, Augsburg, Italy, Yellow River).
- Achieved state-of-the-art quantitative and qualitative results compared to existing classification methods.
- The dynamic scan module effectively captured crucial spectral-spatial details, and the fusion rule successfully merged heterogeneous features.
Conclusions:
- The SDSM method offers a significant advancement in multi-source remote sensing data classification by enhancing feature extraction and fusion.
- The adaptive nature of the dynamic scan module and the guided fusion strategy are key to the method's success.
- The findings suggest a promising direction for leveraging diverse remote sensing data through sophisticated deep learning architectures.
