Sensors & Transducers (Oct 2015)
An Empirical Study for Quantification of Carcinogenic Formaldehyde by Integrating a Probabilistic Framework with Spike Latency Patterns in an Electronic Nose
Abstract
Recently, exposure to formaldehyde has appeared as a major concern since it has been listed as a human carcinogen. Conventional methods for its long-term monitoring are not feasible due to their high operational cost, long analysis time and the requirement of specialized equipment and staff. In this paper, we develop an electronic nose, containing an array of commercially available low cost Figaro gas sensors, to support autonomous and long-term monitoring of formaldehyde. Hardware friendly gas quantification without requiring any manual tuning of parameters is the major challenge with the electronic nose. We handle this challenge by treating it as a classification problem because data acquisition at continuously varying concentrations may incur large expense and a great deal of time. Instead, twenty different concentrations of formaldehyde with 0.25 ppm increment step in the target range between 0.25 to 5 ppm, spanning commonly found formaldehyde levels in indoor and outdoor environments, are input to obtain its signatures in order to quantify/classify its levels within this target range. A computationally efficient bio-inspired spike latency coding scheme, in which spike latencies corresponding to sensitivity patterns of the sensors in the array shift with the change in concentration, is targeted for this purpose. However, stochastic variability in the spike latency patterns, corresponding to repeated exposure to the same formaldehyde concentration level, is observed. We target two Bayesian inference methods, namely multivariate Bayesian and naive Bayes, to express the uncertainty about the spike latency patterns in terms of a probability encoding framework. These methods do not require any manual tuning of parameters, in contrast to other state of the art methods. A best performance of 95.75 % is achieved with the naive Bayes method on the experimentally obtained data set of formaldehyde.
