A multi-module artificial neural network approach to pattern recognition with optimized nanostructured sensor array

The selection of appropriate sensing array nanomaterials and the pattern recognition of sensing signals are two challenges for the development of sensitive, selective, and cost-effective sensor array systems. To tackle both challenges, the work described in this paper focuses on the development of a new hybrid method which couples multi-module method with artificial neural networks (ANNs) for the optimization-optimized multi-module ANN classifier (OMAC) to enhance the correct detection rate for multiple volatile organic compounds (VOCs). In this OMAC method, each module is dedicated to a group of VOCs with specific inputs. Each sensor element's selectivity is quantitatively evaluated to assist the selection of sensing array materials, which also facilitates the selection of inputs to each dedicated neural network module. This OMAC method is shown to be useful for achieving a high overall recognition rate for a selected set of vapor analytes. The results are discussed, along with the implications to the better design of ANN pattern classifiers in chemical sensor applications.