A Double Triangular Feature-Based Sensor Sequence Coding Approach for Identifying Chinese Liquors Using an E-Nose System

Herein, a novel model-free double triangular feature-based sensor sequence coding approach is developed to identify Chinese liquors using a self-designed electronic nose (e-nose) system. In the feature extraction stage, the response curve of each gas sensor of the e-nose is averagely divided into ten segments, followed by a summation of the response voltages in each segment to form the voltage features. Two triangular feature sets, of which the number of the feature layers is equal to the number of sensors used, are subsequently constructed for each segment of each liquor sample based on voltage features. The feature’s sensor sequence (SS) codes are generated by arranging features corresponding to different sensors or different positions of the same layer in each triangular feature set in ascending order. The sample SS codes are obtained when all the feature SS codes within a sample are connected. The SS codes of each liquor type are then constructed by integrating the coincident SS codes with the inconsistent SS codes of all the training samples within the liquor type, where the latter are marked as zero. In the identification stage, the liquor type whose SS codes are the most similar to those of the testing samples is selected as the predicted result of the sample. The proposed method has an average accuracy of 96.3%, which is considerably higher than traditional classification algorithms, such as random forest, support vector machine, naive Bayesian, k nearest neighbor, voting-extreme learning machine, long short-term memory, and gated recurrent unit.