查看文章

We report on the building of a simple and reproducible electronic nose based on commercially available metal oxide gas sensors aimed at monitoring the freshness of sardines stored at 4°C. Sample delivery is based on the dynamic headspace method and four features are extracted from the transient response of each sensor. By using an unsupervised method, namely principal component analysis (PCA), we found that sardine samples could be grouped into three categories (fresh, medium and aged), which corresponded to an increasing number of days that sardines had spent under cold storage. Then, supervised linear or non-linear pattern recognition methods (PARC) such as discriminant factor analysis (DFA) or fuzzy ARTMAP neural networks (FANN) were successfully applied to build classification models to sort sardine samples according to these three states of freshness. The success rate in classification …