Analysis of transient signals by feature extraction from time-frequency images

Abstract

Deep analysis of transient and complex behavior signals can be solved by time — frequency transforms based methods. Examples come from thermal, vibration and high-voltage circuits and applications. After applying suitable transform to the analyzed signals an image is obtained, with relevant values concentrated in some regions. This aspect is known as energy concentration, which depends on the source of the signal and on the channel of the propagation. The region of interest must be selected and separately analyzed in order to extract the relevant information about the signal. The work promotes two approaches for this analysis: (i) by using time-frequency transform followed by region selection and feature extraction, and (ii) analysis of the information content, by using Renyi entropy. The key in solving the problem of analysis is to extract a finite and relevant set offeatures from the time-frequency image. The analysis from information point of view via Renyi entropy allows evaluating the signal complexity, in terms of the components number and basic properties of each component. The preliminary obtained results motivate us to continue the analysis and to develop new algorithms for the analysis of such signals.