Adaptive Remote Sensing Image Enhancement for KOMPSAT Imagery

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Satellite imagery, such as KOMPSAT, faces quality degradation from atmospheric effects, low illumination, and viewing angles.
• Manual image enhancement is labor-intensive and inconsistent, hindering accurate deep learning-based segmentation.
• Degraded imagery significantly reduces the performance of deep learning models for image segmentation tasks.

Purpose of the Study:

• To develop an automated framework for enhancing degraded remote sensing imagery.
• To improve the segmentation performance of deep learning models on challenging satellite data.
• To introduce a reinforcement learning-based approach for adaptive image enhancement.

Main Methods:

• Proposed Adaptive Remote Sensing Image Enhancement (ARSIE), a reinforcement learning-based framework.
• ARSIE learns image-specific enhancement sequences from a filter pool using a policy network.
• The policy network leverages intermediate feature maps from a segmentation model to guide enhancement decisions.

Main Results:

• ARSIE automatically discovers effective, image-specific enhancement combinations.
• Consistent improvements in segmentation accuracy were observed on degraded KOMPSAT imagery.
• The framework demonstrated the ability to enhance image quality for downstream tasks.

Conclusions:

• ARSIE offers an automated and effective solution for enhancing degraded remote sensing images.
• The reinforcement learning approach directly optimizes for improved segmentation performance.
• ARSIE shows potential for broader application in improving various types of satellite imagery quality.