Since 2023, a central sensor on HiRISE has been malfunctioning intermittently, depriving the instrument of half its colour images. HiPredict fills the gap with an error margin of 1.1 per cent.
The HiRISE camera on the Mars Reconnaissance Orbiter spacecraft photographs the Martian surface at 25-30 cm per pixel. Its 14 detectors partially overlap, making it possible to produce infrared color images that are valuable for spotting compositional variations and active processes. But since 2023, the central RED4 detector has been intermittently failing, which has halved the instrument's color capacity.
More than just scaling
The official processing chain replaces this missing channel with a synthetic channel, obtained by linearly scaling the neighboring infrared channel — an approach that is unable to account for nonlinear color relationships. Instead, HiPredict, a U-Net network, combines the co-located infrared and blue-green channels, neighboring red observations, and acquisition metadata. Trained on 81,717 samples from 13,625 observations, it reconstructs RED4 with a normalized average absolute error of 1.1% over 23,348 test samples, compared to 3.1% for the current method.
The reconstructed images more faithfully reproduce structures sought by geologists: icy craters, exposed strata on ridges or chloride deposits. The tool is published in free software, and its authors see it as a way to extend the scientific life of missions whose instruments are aging.
Read the full article on the original source
Go to the source →