Ensemble model with cascade attention mechanism for high-resolution remote sensing image scene classification.

Area of Science:

• Earth Observation
• Computer Vision
• Machine Learning

Background:

• High-resolution remote sensing image classification is crucial for Earth observation.
• Existing methods struggle with limited labeled data and over-reliance on global information.
• Class-specific ground objects in local regions are key to accurate image categorization.

Purpose of the Study:

• To develop an improved model for high-resolution remote sensing image scene classification.
• To address limitations of existing models regarding data scarcity and global information dependency.
• To enhance the model's ability to focus on discriminative regions within images.

Main Methods:

• An ensemble model featuring a cascade attention mechanism and two convolutional neural network (CNN) branches was proposed.
• Each CNN branch was trained on diverse large datasets to improve feature extractor generality and prior knowledge.
• A cascade attention mechanism was developed to integrate branches and capture the most discriminative information.

Main Results:

• The proposed cascade attention-based double branches model achieved state-of-the-art performance.
• Experiments were conducted on four benchmark datasets: OPTIMAL-31, UC Merced Land-Use Dataset, Aerial Image Dataset, and NWPU-RESISC45.
• The model demonstrated superior performance across all tested benchmark datasets.

Conclusions:

• The developed end-to-end model effectively addresses challenges in high-resolution remote sensing image classification.
• The cascade attention mechanism successfully guides the model to focus on class-specific regions.
• The approach offers a significant advancement in the field of Earth observation and image analysis.