Classification of fresh edible oils using a coated piezoelectric sensor array-based electronic nose with soft computing approach for pattern recognition

An electronic nose based on an array of six bulk acoustic wave polymer-coated piezoelectric quartz (PZQ) sensors with soft computing-based pattern recognition was used for the classification of edible oils. The electronic nose was presented with 346 samples of fresh edible oil headspace volatiles, generated at 45degreesC. Extra virgin olive (EVO), nonvirgin olive oil (NVO) and sunflower oil (SFO) were used over a period of 30 days. The sensor responses were visualized by plotting the results from principal component analysis (PCA). Classification of edible oils was carried out using fuzzy c-means as well as radial basis function (RBF) neural networks both from a raw data and data after having been preprocessed by fuzzy c-means. The fuzzy c-means results were poor (74%) due to the different cluster sizes. The result of RBF with fuzzy c-means preprocessing was 95% and 99% for raw data input. RBF networks with fuzzy c-means preprocessing provide the advantage of a simple architecture that is quicker to train.