HiPredict: Predicting the missing HiRISE RED4 spectral channel with machine learning
HiPredict’s U-Net reconstructs failed HiRISE RED4 imagery at 1.1% error, beating linear SYN4.
HiRISE on the Mars Reconnaissance Orbiter lost reliable RED4 colour-infrared coverage after a central detector began failing intermittently in 2023. The existing pipeline synthesizes RED4 by linearly scaling the co-located IR10 channel, which misses nonlinear relationships. HiPredict, a U-Net, reconstructs RED4 from IR10, BG12, neighbouring red observations, and metadata, trained on 81,717 samples from 13,625 observations. On 23,348 test samples from 3,892 pre-failure observations it reaches 1.1% normalised mean absolute error, versus 3.1% for SYN4, and is released as open software.
- HiRISE RED4 began failing intermittently in 2023, cutting colour imaging capability in half.
- HiPredict is a U-Net using IR10, BG12, neighbouring red data, and acquisition metadata.
- Trained on 81,717 samples from 13,625 observations.
- Test NMAE is 1.1% versus 3.1% for the linear SYN4 baseline; the tool is open source.
From the original article
1 min read at arxiv.orgThe High Resolution Imaging Science Experiment (HiRISE) onboard the Mars Reconnaissance Orbiter (MRO) acquires imagery of Mars at 25--30 cm/pixel, supporting studies of martian geology, climate history, and landing-site characterization.
By Ardan Suphi · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.