A signal denoising technique based on wavelets modulus maxima lines and a self-scalable grid classifier

• Published in: 2015 IEEE Workshop on Signal Processing Systems (SiPS) pp. 1 - 6
• Main Authors: Vasconcelos Machado, Rubem Geraldo, de Oliveira Mota, Hilton
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2015
• Subjects: denoising; multiple-instance learning; Neurons; Noise reduction; Partial discharges; self-organizing maps; Training; Transient analysis; Wavelet transforms; wavelets transform
• DOI: 10.1109/SiPS.2015.7344974

This paper presents the description of a signal processing technique using the Wavelets Transform and a self-scalable grid classifier. The procedure is based on the cycle-spinning approach applied to the Translation-invariant Wavelet Transform. It exploits the characteristics of the Wavelets modulus maxima propagation along decomposition levels (scales) as the criterion to select the relevant coefficients. Selection was performed by a data classifier inspired on a Self-organizing Map but with enhancements to incorporate self-scalability and multiple instance learning capabilities. The procedure was employed for the processing of Partial Discharge signals, which is a technique for the diagnostics of high-voltage equipment. We performed comparisons with standard form classifiers based on the Multilayer Perceptron and Support Vector Machines. The results show that the technique allows the same orders of accuracy and generalization of those classifiers, but with the advantages of self-scalability, dimensional independence, low processing cost and high degree of parallelization.