The Advantage of Multispectral Images in Fruit Quality Control for Extra Virgin Olive Oil Production

Abstract

The highest quality of extra virgin olive oil must be guaranteed to succeed in the competitive market of the industry. To fulfil this need, high standards of fruit quality are required. In recent years, this search has led to the development of computer vision systems for the automatic inspection of fruits. Skin damage, dirtiness, and external colour were the main features selected. However, no known report is related to the internal quality evaluation of olives. The present research explores the application of multispectral images (from 600 to 975 nm) to discriminate olive fruits by their firmness. To achieve this aim, 58 olives were classified as hard or soft by an expert and then measured by a penetrometer. Afterwards, 58 multispectral images were obtained, and 8352 pixels were randomly selected for feature extraction. The feature vector of each pixel was composed of 25 infrared absorption values. These data were input into three supervised classification algorithms: a naïve Bayes (NB) classifier, a multilayer perceptron (MLP), and a support vector machine variation (SMO). The highest performance was achieved by the MLP, with an accuracy of 95% for the firmness grading. Furthermore, the three algorithms were subjected to principal component analysis (PCA), by which the MLP was confirmed to be the best model and its results were improved. Moreover, according to the second principal component, the 675 nm and 855 nm band frequencies have the strongest weights. The overall results confirm the convenience of multispectral imaging in olive firmness inspection, with the advantage of being a non-invasive technology.