A benchmarking protocol for SAR colorization: From regression to deep learning approaches.

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Synthetic Aperture Radar (SAR) images are crucial for remote sensing but are difficult to interpret due to inherent speckle noise and grayscale limitations.
• SAR image colorization aims to enhance interpretability by adding color while preserving spatial and radiometric information, though the field is nascent.

Purpose of the Study:

• To establish a comprehensive research line for supervised learning-based SAR image colorization.
• To introduce a protocol for generating synthetic color SAR images and propose novel assessment metrics.
• To present and evaluate an effective conditional generative adversarial network (cGAN) based method for SAR colorization.

Main Methods:

• Development of a protocol for synthetic color SAR image generation.
• Implementation of baseline methods for SAR colorization.
• Design and application of a conditional generative adversarial network (cGAN) for SAR image colorization.
• Proposal of new numerical assessment metrics tailored for SAR colorization.

Main Results:

• The proposed cGAN-based network demonstrates significant effectiveness in SAR image colorization.
• Extensive testing validates the performance of the developed SAR colorization approach.
• The research provides a benchmark and evaluation protocol for future SAR colorization studies.

Conclusions:

• The proposed research line, including the protocol, benchmark, and cGAN method, represents a significant advancement in SAR image colorization.
• The developed cGAN-based approach effectively addresses the challenges of speckle noise and grayscale nature in SAR imagery.
• This work lays the foundation for further research and development in automated SAR image interpretation.