Dimensionality Reduction In Computationally Inteligent Photoacoustic Measurement Data Processing

Abstract

The research presented in this paper is part of an effort to improve the photoacoustics measurement signal autocorrection method, based on computational intelligence. Autocorrection here deals only with the elimination of distortion of the experimental signal due to the influence of the measuring system. The method implies two already developed models - regression and classification, which in cooperation recognize the transfer function of the microphone in the frequency domain, as an accurate representation of the microphone response. The obtained data are significant for the signal correction procedure, the result of which is an undistorted signal, used in the process of material characterization, for accurate, precise, and reliable determination of the parameters of the tested sample. Further research goes in the direction of dimensionality reduction for both regression and classification models without losing the quality of measurements. This paper presents the results of classification model improvement. Namely, testing the model under different conditions (theoretical or experimental signals, with and without noise, different types of microphones, different samples) we found that the accuracy of the model is high and that the processing speed of measured data does not change seriously by reducing the number of measurement points and thus reducing dimension of the models’ input vector. Principal component analysis, discussion of feature correlations and expert knowledge were used to determine the number and measurement points frequency. It was proved that the procedure of measuring and classification of microphones can be performed simply and quickly by measuring at 1 defined point.