Quantile-Frequency Analysis and Deep Learning for Signal Classification

Abstract

This paper proposes a new method for signal classification based on a combination of recently introduced nonlinear spectral analysis technique called quantile-frequency analysis (QFA) and deep-learning (DL) image classifiers. The QFA method converts a one-dimensional signal into a two-dimensional representation of the signal's oscillatory behavior in the frequency domain at different quantiles or a sequence of such representations that are localized in time. The DL image classifiers utilize these representations for signal classification. The benefit of this QFA-DL classification method, especially in comparison with the traditional method based on the power spectrum and spectrogram, is demonstrated by a numerical experiment using real-world data from a nondestructive evaluation (NDE) application.

Local Gaussian Cross-Spectrum Analysis

“…In many cases, it can be difficult to interpret the various spectra/visualizations which result from these modifications of the classical approach, but machine learning methods can be combined with these techniques and help reduce/remove this problem-see, e.g., Chen et al (2021); Li (2020Li ( , 2022b for some examples of this approach. The review paper Ciaburro and Iannace (2021) contains more information about machine-learning-based methods applied to time series data.…”