Coastal Land Cover Classification of High-Resolution Remote Sensing Images Using Attention-Driven Context Encoding...

Area of Science:

• Remote Sensing
• Computer Vision
• Geospatial Analysis

Background:

• Coastal land cover classification (CLCC) from high-resolution remote sensing images is challenging due to low inter-class variance and complex spatial details.
• Fully convolutional neural networks (FCNNs) are common but limited by small receptive fields, hindering local context capture.
• Existing complex decoders in FCNNs can introduce redundancy and increase computational load.

Purpose of the Study:

• To propose a novel attention-driven context encoding network to address limitations in CLCC.
• To enhance the capture of multi-scale spatial details and global context in remote sensing images.
• To improve the accuracy and efficiency of coastal land cover classification.

Main Methods:

• Developed a novel attention-driven context encoding network incorporating lightweight global feature attention modules.
• Integrated position and channel attention modules to capture long-range dependencies and multi-dimensional global context.
• Employed multiple objective functions for supervised optimization of scale-specific feature information.

Main Results:

• The proposed method demonstrated optimal performance in encoding long-range context and recognizing spatial details.
• Achieved superior evaluation index representations compared to state-of-the-art approaches in CLCC tasks.
• Effectively aggregated multi-scale spatial details during the decoding stage.

Conclusions:

• The attention-driven network significantly enhances feature representation for CLCC.
• The method overcomes the limitations of traditional FCNNs in capturing global context and spatial details.
• The proposed approach offers an effective solution for accurate and efficient coastal land cover classification.