SAR Imagery Colorization.
A deep-learning model that colours grayscale synthetic-aperture radar, so surface features are easier to read.
- SAR → RGB
- Grayscale radar in, colour out
- Paired
- Learned from SAR and optical pairs
- PSNR
- Measured, not judged by eye
The idea
Synthetic aperture radar carries rich structural and textural information — it is used from geology to environmental monitoring — but it arrives in grayscale, which makes it hard to read at a glance.
The approach
A deep-learning model learns colour from paired SAR and optical images, trained to minimise a custom loss between the predicted and the true colour image, so surface features become distinguishable.
The hard parts
SAR is inherently noisy, so preprocessing matters; the architecture has to respect the statistics of radar rather than photographs; and quality is evaluated with metrics such as PSNR.
Keystone