Remote Sensing Scene Classification via Multi-Branch Local Attention Network.

Area of Science:

• Computer Science
• Geoscience

Background:

• Remote sensing scene classification (RSSC) is crucial for image interpretation.
• Convolutional Neural Networks (CNNs) have advanced RSSC, but struggle with inter-class similarity and intra-class diversity.
• Existing CNNs often exhibit insufficient differentiation in extracted feature representations.

Purpose of the Study:

• To propose an effective method for remote sensing image scene classification.
• To address the limitations of insufficient feature representation differentiation in CNNs for RSSC.
• To enhance the accuracy and robustness of remote sensing scene classification.

Main Methods:

• A novel Multi-Branch Local Attention Network (MBLANet) is proposed for RSSC.
• The MBLANet integrates a Convolutional Local Attention Module (CLAM) into a ResNet backbone.
• CLAM comprises parallel Convolutional Channel Attention Module (CCAM) and Local Spatial Attention Module (LSAM) for improved feature representation.

Main Results:

• Extensive experiments were conducted on three benchmark datasets.
• The proposed MBLANet demonstrated superior performance compared to state-of-the-art methods.
• The integration of CLAM effectively improved feature representation capabilities.

Conclusions:

• The MBLANet method significantly advances remote sensing scene classification.
• The proposed attention mechanism enhances the ability to emphasize main targets in complex backgrounds.
• MBLANet offers a promising solution for accurate and robust remote sensing image interpretation.