Antarctic Ice Core Analysis

This project used an interesting, unconventional application for eCognition in order to determine correlation to the spatial pattern of ice grains to explain differences in their formation through time. The image is a down-scaled version of a slide of ice crystal grains from an ice core (cut longitudinally).

This is the base image; the colors in the image have no meaning, the only feature of interest the size and shape of the crystal grains. 

The green line in the center represents a change in electrical response, the objective is to see if there's a difference in grain structure related to the difference in response.

The first step was to separate the two sides of the image, taking care to separate out the crystals in the transition zone.

The second step was to remove all crystals with shape characteristics altered by the boundary of the image.

I used a Random Forest classifier to help find the shape characteristics that maximize separation between the groups. 

The Border Index feature is the single most important separator between the two groups. 

Crystals above the mean value of the Border Index are highlighted in yellow to visualize the maximum geometric difference between the charge groups.