Optical Nose Signal Denoising and Detrending Using Wavelets

摘要

Sensor miniaturization and the drive to develop ultra compact chemical sensing devices have received much attention in recent years. The characteristics, nature, and quality of signals are inevitably affected as sensors reduce in size. We have developed micron-size sensing elements using solvatochromic compounds and porous silica microspheres. Although many sensor elements, depending on their chemistry, show strong signals, a number of sensors demonstrate substantial drift and noise in their responses. The signals obtained from these miniaturized sensors comprise three dominant components: 1) the main component, corresponding to the response of the volatile of interest, 2) a piecewise linear trend, due to the delivery system and sensor recovery, and 3) a random noise component. While noise treatment can be done using many well-known methods, e.g., Fourier-based digital filtering, the treatment of the trend component is rather difficult. The objective of this study is to reduce the effect of the undesirable components of the time response, i.e., noise and trend, in the presence of weak signals. Trends are due to the non-stationary structure of the time-series signals, and the pulsatory switching of the volatiles in a nose system. Ad hoc methods of trend removal cause unsatisfactory results, by introducing severe discontinuities. In recent years, wavelet transform has become an invaluable tool for the treatment of non-stationary signals. In this study, the detrending and denoising is performed in one step, using wavelet transform. It is illustrated that for the optical-nose (O-Nose) signals, this method of preprocessing removes the original trend of the signals, while introducing no artificial aberrations.