Fragments from splitting atoms leave tiny damage tracks in minerals. Geologists chemically etch the sample to expose those marks, then count them under a microscope as one input to dating rocks and reconstructing their heating history.[1]
HALtracks2D identifies tracks in images of apatite and mica. It uses light passing through the sample and light reflected from its surface, helping connect a surface opening to the underlying track.[1]
In one comparison, its mean track-density estimates differed from an experienced analyst’s by about 4% on grains without non-track features and 20% on grains with them. The analyst had already excluded heavily cluttered grains, judged to have a noise-to-signal ratio above 0.5. Here, noise means marks judged not to be tracks.[1]
The result measures counting agreement on selected grains. It does not independently validate geological ages, and the hardest images remain outside this comparison.[1]
Sources & further reading
- OK computer: AI reaches human-level accuracy in fission track counting in apatite and mica Earth and Planetary Science Letters / author’s copy · Accessed 20 Oct 2026