An electronic nose (“e-nose”) is a sensor used to selectively measure volatile organic compounds. Although e-noses have advantages in terms of cost and ease of use, they also have inherent limitations in terms of sensitivity to detect subtle variations in compound concentrations, leading to inconsistent results if not properly managed. The data generated by e noses generally require advanced processing techniques for interpretation of complex signal patterns. This is why e-nose food classification applications tend to use Deep Learning techniques such as Recurrent Neural Networks.
In this publication (open access) the authors used an array of 7 sensors to build a model to differentiate pork, bovine and fish gelatin. The model was based on a commercial sample of each, dissolved in water as a 1% solution and warmed. The model was then applied to different in-house mixtures of the gelatins at different time-points after preparation. The authors do no report if it was validated with orthogonal samples of verified origin. The sensors had selective sensitivity to a range of volatiles including ethanol, methane, propane, butane, ammonia and hydrogen sulfide.
The authors report that classification efficiency, as measured by the AUC (Area Under the ROC Curve), was variable when considering one sensor in isolation but was good when all 7 sensors were multiplexed. The AUC increased with time from sample preparation, rising to over 98% at 2-hours from the samples being prepared. The authors conclude that this makes the technique a promising candidate for constructing a routine instrument to check the species of commercial gelatin.