Probabilistic Rank Score Coding: A Robust Rank-Order Based Classifier for Electronic Nose Applications

Motivated by the recent experimental findings about odor identification with the unique spiking patterns of neurons in the biological olfactory system, rank-order-based classifiers have been proposed for gas identification in electronic nose applications. These classifiers rely on one-to-one mapping between the target gas and the temporal sequence of spiking sensors in an electronic nose. However, shuffled spike sequences, due to low repeatability of the response patterns from the sensors, limit the performance of these classifiers. We propose a robust probabilistic rank score coding scheme that tabulates the probability of each spiking sensor at each rank, or temporal position, by exploiting all the spike sequences of the sensors for each target gas, and we analyze a new test vector with this tabular information for its identification. A new quantification metric is proposed in order to estimate the confidence of the classifier's output. Overall, our coding scheme provides an analytical solution that spots the most probable gas for any new test spike sequence with the quantitative feedback. In order to evaluate the robustness of our coding scheme, we target the identification of commonly found health-endangering indoor gases, such as C6H6, CH2O, CO, NO2, and SO2, as a case study. Data of these gases are acquired using our in-house fabricated gas sensor array under different operating conditions, as well as an array of commercially available gas sensors under uniform operational settings. An accuracy rate of 100% has been achieved in both cases with our probabilistic rank score coding scheme.