Abstract

Unbiased internal quality classification of Chestnuts (Castanea spp.) is extremely important to the fresh and processed industries. It can also be used as a tool for applied scientific studies, such as the training of non-invasive techniques to determine chestnut internal quality, and the effect of pre- and post-harvest treatments. At the moment, humans visually perform the invasive quality assessment of chestnuts. This procedure is prone to errors and high variability due to individuals' fatigue, lack of training, and subjectivity. Thus, there is a need to develop a technique that is able to objectively classify internal quality of chestnuts. In this paper, a computer vision methodology is proposed to sort chestnuts into five classes, as established by an expert human rater. 1790 color images from slices with different quality classes were acquired, using a flat panel scanner, from the hybrid cultivar 'Colossal' and 'Chinese seedlings'. After preprocessing, a total of 1931 color, textural, and geometric features were extracted from each color image. Furthermore, the most relevant features were selected using a sequential forward selection algorithm. Thirty-six features were found to be effective in designing a quadratic discriminant classifier with a cross-validated overall performance accuracy of 89.6%. These results showed that this method is an accurate, reliable, and objective tool to determine chestnut slice quality, and might be applicable to in-line sorting systems. (C) 2012 Elsevier Ltd. All rights reserved.