Adaptive K-NN for the Detection of Air Pollutants With a Sensor Array

Abstract

The field of air-quality monitoring is gaining increasing interest, with regard to both indoor environment and air-pollution control in open space. This work introduces a pattern recognition technique based on adaptive K-nn applied to a multisensor system, optimized for the recognition of some relevant tracers for air pollution in outdoor environment, namely benzene, toluene, and xylene (BTX), NO/sub 2/, and CO. The pattern-recognition technique employed aims at recognizing the target gases within an air sample of unknown composition and at estimating their concentrations. It is based on PCA and K-nn classification with an adaptive vote technique based on the gas concentrations of the training samples associated to the K-neighbors. The system is tested in a controlled environment composed of synthetic air with a fixed humidity rate (30%) at concentrations in the ppm range for BTX and NO/sub 2/, in the range of 10 ppm for CO. The pattern recognition technique is experimented on a knowledge base composed of a limited number of samples (130), with the adoption of a leave-one-out procedure in order to estimate the classification probability. In these conditions, the system demonstrates the capability to recognize the presence of the target gases in controlled conditions with a high hit-rate. Moreover, the concentrations of the individual components of the test samples are successfully estimated for BTX and NO/sub 2/ in more than 80% of the considered cases, while a lower hit-rate (69%) is reached for CO.