Preprint Proposes ML Scan of HiRISE Imagery for Artificial Surface Patterns on Mars and Moon

The work reframes technosignature searches from radio signals to durable surface modifications that could persist on airless bodies for geological timescales. It pairs existing 25–30 cm/pixel HiRISE coverage of Mars with emerging anomaly-detection networks, arguing that LiDAR and ground-penetrating radar follow-up could place the first quantitative upper limits on artificial terrain alteration. No detections are claimed; the paper is strictly a methodological prospectus.

Existing SETI efforts have concentrated on electromagnetic leakage or megastructures, yet surface artifacts on geologically dead worlds represent a distinct, long-lived category. The proposal correctly notes that human monuments and mining scars remain visible for millions of years; extending the same logic to Mars requires rigorous controls for fractal geology and impact gardening. Unsupervised clustering risks conflating these natural features with artifacts unless trained on extensive negative examples from terrestrial analog sites.

A key gap is the absence of any quantified false-positive rate or power analysis for the proposed classifiers. Supervised models would need large, verified training sets of both artificial and natural terrain that do not yet exist at planetary scale. Complementary spectroscopy could distinguish processed materials, but the paper does not model integration costs or data volume.

Next steps include pilot application to well-mapped regions such as the lunar south pole and Jezero crater, with results released as open anomaly catalogs for community vetting.