Quality Control of Food Products using Image Analysis and Multivariate Statistical Tools

Abstract

Visual appearance is an important feature in the quality assessment of food products, since it plays a key role in the decisions made by consumers. Frequently, its evaluation is carried out by a panel of experts from the quality department who analyze, by visual inspection, samples of product taken from the process. It is well-known that such methodologies of assessment suffer from several drawbacks, such as subjectivity, limited precision, and lack of stability over time, even for well-trained personnel, although extensive training programs can improve assessment performance. In this context, we present in this paper results regarding the development of an automated methodology for assessing the visual appearance of cereal flakes, in what concerns a particular quality feature, relative to the existence of regions where cereal coating is inadequate. The proposed procedure is able to extract the necessary information from images taken from product samples, leading to an objective, stable, and quantitative quality control measurement system for this property. The developed algorithm essentially consists of implementing a supervised classification methodology, based on the estimated Mahalanobis distance for assessing proximity in the color space, while incorporating the natural variability and color correlations found in cereal flakes. This procedure also integrates fuzzy logic reasoning for samples falling under regions closer to the classes’ boundaries. Results obtained from a real industrial plant indicate that it is indeed possible to classify different samples of flakes according to classes previously defined. They also provide a sound basis for further developments, in particular regarding the generation of metrics for quality assessment and the implementation of a similar procedure online and in situ.