Sensor Array Optimization for Complexity Reduction in Electronic Nose System

Abstract

An Electronic nose (E-Nose) can be used to assess food quality and fruit ripeness without personal bias. A set of relevant sensors must be identified to design an effective E-Nose and reduce implementation cost and complexity. The analysis of tropical fruit odour in terms of pattern recognition errors is carried out to determine the minimum number of sensors and their combinations. Two new methods namely 1) principal component loading and mutual information between sensor data, and 2) threshold based approach are proposed in this work to evaluate and optimize the sensor set. Four pattern recognition methods, namely multilayer perceptron neural network (MLPNN), radial basis function neural network (RBFNN), support vector machine (SVM), and k-nearest neighbour (k-NN) are also compared in terms of classification performance. The pattern recognition error of SVM with the optimal set of sensors is as low as 2.78% and that of k-NN is 9.72%. The results conclude that the pattern classification error with MLPNN, and RBFNN is higher than the error from k-NN and SVM.