NIR detection of grain weevils inside wheat kernels

This paper reports research on the detection of grain weevils growing inside cereal grains. Grains are inspected in the near infrared (NIR) and bright patches characteristic of internal infestation are examined. The best technique for detecting these bright patches is found to be a self-normalisation operation in which a filtered version of the grain image is subtracted from the original. The patches are then sought using weighted averaging templates and the most significant peak is classified. Classification accuracy is around 85% rather higher than the best achieved by human inspectors and far more consistent. This classification accuracy applies to individual grains, but with application of a novel optimisation procedure, batch infestation can be detected with over 95% accuracy: this is high enough to be practically useful, and considerably less costly and more convenient than the alternative of X-ray imaging.