Authors: N.J. Kusabs, L. Trigg, A.F. Bollen, G. Holmes
Addresses: Supply Chain Systems Group, Lincoln Ventures Ltd, Ruakura Research Centre, Hamilton, New Zealand. ' Reel Two Incorporated, Hamilton, New Zealand. ' Supply Chain Systems Group, Lincoln Ventures Ltd, Ruakura Research Centre, Hamilton, New Zealand. ' Machine Learning Group, University of Waikato, Hamilton, New Zealand
Abstract: A machine learning classification system has been developed to sort mushrooms into similar quality grades to those used by human inspectors. The attributes considered for the algorithm included weight, firmness, image features and some subjective scales of the common cultivated button mushroom. Two grading systems were tested; one involving three broad quality grades and a more detailed five-grade system. Two machine learning methods were used to build quality prediction models, relative to the mushroom quality grade criteria specified by the inspectors. A head inspector was used as a reference grading expert and the mis-classification error of the models based on his/her grading for both three and five grades varied from 17 to 22%. The accuracy of the quality prediction models is at the upper level of the variation measured between mushroom industry inspectors. This machine learning classification system provided insights into the subjective decision-making regarding mushroom quality.
Keywords: agaricus bisporus; quality grading; machine learning; mushroom quality; quality prediction; visual analysis; automated inspection; button mushrooms; weight; firmness; image features; decision making; human inspectors; postharvest.
International Journal of Postharvest Technology and Innovation, 2006 Vol.1 No.2, pp.189 - 201
Published online: 11 Dec 2006 *Full-text access for editors Access for subscribers Purchase this article Comment on this article