Sparse Focus Network for Multi-Source Remote Sensing Data Classification

Xuepeng Jin,Junyan Lin,Feng Gao,Lin Qi,Yang Zhou

2024-06-03

Abstract:Multi-source remote sensing data classification has emerged as a prominent research topic with the advancement of various sensors. Existing multi-source data classification methods are susceptible to irrelevant information interference during multi-source feature extraction and fusion. To solve this issue, we propose a sparse focus network for multi-source data classification. Sparse attention is employed in Transformer block for HSI and SAR/LiDAR feature extraction, thereby the most useful self-attention values are maintained for better feature aggregation. Furthermore, cross-attention is used to enhance multi-source feature interactions, and further improves the efficiency of cross-modal feature fusion. Experimental results on the Berlin and Houston2018 datasets highlight the effectiveness of SF-Net, outperforming existing state-of-the-art methods.

Image and Video Processing

What problem does this paper attempt to address?

The paper mainly addresses the issue of multi-source remote sensing data classification, particularly the challenges encountered when processing hyperspectral images (HSI), synthetic aperture radar (SAR), and LiDAR data. Existing multi-source data classification methods are easily affected by irrelevant information during multi-source feature extraction and fusion, leading to compromised classification performance.

To solve this problem, the authors propose a new method called Sparse Focus Network (SF-Net). This method optimizes the feature interaction process of the model by introducing a sparse attention mechanism, reducing the interference caused by irrelevant features, and retaining only the most relevant self-attention values to improve feature aggregation efficiency. Specifically, SF-Net uses a sparse attention mechanism to extract HSI and SAR/LiDAR features and employs cross-attention to further enhance the interaction between multi-source features, thereby improving the efficiency of cross-modal feature fusion.

The experimental section demonstrates the performance of SF-Net on the Berlin and Houston2018 datasets. The results show that this method outperforms current advanced methods, exhibiting significant advantages in terms of overall accuracy (OA). This proves the effectiveness of SF-Net in the task of multi-source remote sensing data classification.