The Discrete Lifting Shapelet Transform for Biological Pattern Recognition Ruben-Dario Pinzon-Morales, Member, IEEE,Alvaro-Angel Orozco-Gutierrez,
Cesar-German Castellanos, Rodrigo Capobianco Guido · 2011
Wavelet transform has been widely used in bio- logical signal processing since last decade. The success in the wavelet results relies on the proper selection of the mother wavelet function. In this document the wavelet function is customized to the application. Mentioned approach is possible by means of the Discrete Lifting Shapelet Transform (DLST), a novel transform introduced here and use to generate mother wavelets that resemble the shape of a match pattern and the time-frequency information within. The DLST is inspired on two works: the Discrete Shapelet Transform and the Signal- Dependent Filter Banks. Comparison results are given ensuring the efficacy of the proposed transform in the framework of hand movement recognition using electromyographic signals. I. INTRODUCTION Wavelet analysis is a powerful tool for digital signal processing. It has been widely used in bioelectric signals including evoke-related potentials (ERP), electromyography signals (EMG), microelectrode recordings (MER), electro- cardiogram (ECG), electroencephalograms (EEG), among others. Although there are plenty of wavelet prototypes in the literature, there is not an established rule that states which wavelet may be used for each application. Instead of that, it is an usual task for the researcher to test more o less arbitrarily different wavelet shapes to find one suited. To improve above issue, this article proposed to consider intrinsic information of the signal under analysis, such as the shape, or the frequency dynamic, to construct unique mother wavelet completely customized to the application. Mentioned approach is possible by means of the Discrete Lifting Shapelet Transform (DLST), a novel transform employed to generate mother wavelets that resemble the shape of a match pattern and the time-frequency information within. The DLST is inspired on two works: the Discrete Shapelet Transform and the Signal-Dependent Filter Banks (SDFB). The former described in (1), relates the analytical creation of new mother wavelets that provide and adequate framework to find the time-support of frequencies, and, at the same time, match patterns. The later corresponds to the construction of new mother wavelets based on the input signal itself, which exhibit unique time-frequency response, using lifting schemes and evolutionary techniques (2). Main motivations behind the combination of both approaches are two. Firstly,