Four potato packers add AI grading to their production lines

Ellips announced four North American Elisam potato-grading installations on 2 October. Its True-AI software assesses size, shape, colour and external defects as potatoes move through the line.[1]

The mechanics help make the images useful. Ellips describes rotating carriers that expose the potato’s surface to cameras, with up to 90 images per potato. Rotation gives the classifier views that a single photograph would miss.[2]

The company’s earlier technical account describes deep-learning models trained on hundreds of thousands of reference images, with examples gathered across varieties, seasons and growing conditions. It assigns operators a continuing role in translating customer quality requirements into grading choices.[3]

A grade contains several decisions

USDA’s U.S. No. 1 grade illustrates why that task needs more than an attractive surface. Its criteria include firmness, cleanliness, shape, specified disease and damage exclusions, size and tolerances. The usual minimum diameter is 1⅞ inches unless otherwise specified. U.S. No. 2 uses different shape and damage criteria and a usual minimum of 1½ inches. These are defined combinations of requirements, not positions on one universal quality score.[4]

For an illustrative packing order, suppose a buyer asks for potatoes within a narrow size band. A potato outside that band could still satisfy another buyer’s specification. The useful decision is then to allocate it to a suitable order, while separately checking defects and other requirements. Changing the destination need not mean changing what the camera saw.

Ellips says grading programmes can be adjusted as varieties and customer requirements change. That flexibility connects the machine’s recognition task to the packer’s changing order book.[1]

Report an error or suggest a correction