Hybrid Pattern Recognition for Rapid Explosive Sensing With Comprehensive Analysis

Abstract

This paper presents a hybrid pattern recognition with temperature compensation (HPR-TC) used within an E-Nose system. HPR-TC with E-nose has the novelty, amongst MEMS sensor platforms, of having two modes of operation i.e., rapid mode of detection to be used in time-critical conditions and comprehensive analysis mode for improved detection accuracy. Two modes of operations in HPR-TC are possible because of the implementation of hybrid PR featuring a combination of two different data analysis techniques for explosive sensing. The first part of the hybrid PR is the binary PR based on threshold-based detection and the second one is the analog PR based on PCA and K-mean. The E-Nose system with proposed HPR-TC is validated with two different highly sensitive MEMS sensor types, i.e., SU8 and Si3Nx piezo-resistive cantilever. These MEMS sensors are coated with surface receptors, 4-MBA, 6-MNA and 4-ATP, to improve the selectivity. The E-Nose system can detect explosive compounds such as TNT, RDX, and PETN, in a controlled environment at a concentration as low as 16ppb of TNT, 56ppb of RDX and 134ppb of PETN. Furthermore, measurements show that E-Nose with temperature compensated binary PR can detect the explosives with a detection accuracy higher than 74% as true positives and higher than 79% as true negatives in a short time, within initial 17 seconds of the experiment. However, the temperature compensated analog PR gives a detailed classification of explosives with a higher detection accuracy of 80% as true positives and 86% as true negatives after approximately 95 seconds.