Development of mathematical morphology systems for signal feature extraction and detection

Sun, Pu. (2002) Development of mathematical morphology systems for signal feature extraction and detection. PhD thesis, University of Liverpool.

Text 399060.pdf - Unspecified Download (8MB) | Preview

Abstract

This thesis describes a set of algorithms and systems that were developed, using signal processing techniques based on mathematical morphology (MM), for neonatal electrocardiogram (ECG) signal analysis and power transformer inrush current identification. MM methodologies are founded on set-theoretic concepts and nonlinear superpositions of signals and images. Morphological operations have been applied successfully to a wide range of problems including image processing and analysis tasks, noise suppression, feature extraction and pattern recognition etc. This approach seems very appropriate for dealing with objects which share common features, and has thus attracted attention for solving problems similar to those described in this thesis, which are closely related to feature extraction and identification. This thesis begins with a systematic introduction to MM. It explains the historical background and the concept of MM, highlights the advantages ofMM as an advanced nonlinear signal/image processing technique. A brief comparison between MM and traditional filtering techniques is then given, followed by the descriptions of various morphological operations, from basic operators defined for binary images, to the elaborate generalised framework for sets in a generic mathematical space, the complete lattice. The development of a morphological method to discriminate magnetising inrush current waveform from internal fault conditions of large power transv formers is then described. A morphological signal decomposition scheme is proposed to allow the unique feature associated with the inrush current waveform to be separated and identified in the time domain, to avoid the problems of sensitivity and robustness that may occur in the traditional Fourier analysis based approaches. The performance of the proposed method is assessed and discussed, based on signals derived from various operating conditions of the transformer. The second application presented is a morphological scheme for neonatal ECG signal processing and analysis, aiming to facilitate the investigation of the relationship between the clinical pattern of asphyxiated newborn infants and alterations of the ECG pattern. Neonatal ECGs are not routinely used to achieve a detailed analysis as these measurements would usually involve the time consuming act of manual interpretation and measurement. Existing technologies have also not yet been able to accurately monitor these parameters due to the rapid heart rate and the variation of waveform morphology of babies. In the proposed scheme, a morphological filtering method that incorporates subject specific information is developed, to remove the interferences introduced by recording environments and subjects without much distortion to the ECG pattern of interest. The performance of the proposed algorithms is examined using simulated neonatal ECGs and experimental signals acquired from infants. The possibility of extending this study to the fetuses is also considered, in which the fetal ECG would be obtained from the composite maternal signal, to allow intervention at an early stage for fetuses at a high risk of asphyxia. The implementation and integration of the morphological system for neonatal ECG analysis is then described. A prototype of the morphological ECG analyser is developed, which allows the system to be used in clinics by persons without a detailed knowledge of the technology. The optimisation of basic morphological operators, code design, hardware integration and optimisation are discussed, with emphasis on a generic architecture that can accommodate future improvement and extension without major revision of the code. The results obtained from the pilot trial on the ward of Liverpool Women's Hospital are then given and investigated, focusing on the accuracy of the ECG measurements and the relationship between the waveform morphology and the gestational ages of the babies. The major contributions of this work are the utilisation of the advanced performance of MM for feature enhancement, extraction, noise suppression and background normalisation. The studies include the development of morphological algorithms for the decomposition and representation of the power transformer inrush current waveform, and further to enhance its features of interest and to allow them to be identified; introduction of a novel approach for neonatal ECG signal processing and analysis; development of an integrated morphological system for medical research on the neonatal ECG, and investigation of the results obtained from this system with experiments carried out in a clinical environment.

Item Type:	Thesis (PhD)
Depositing User:	Symplectic Admin
Date Deposited:	20 Oct 2023 15:44
Last Modified:	20 Oct 2023 15:51
DOI:	10.17638/03175097
Copyright Statement:	Copyright © and Moral Rights for this thesis and any accompanying data (where applicable) are retained by the author and/or other copyright owners. A copy can be downloaded for personal non-commercial research or study, without prior permission or charge.
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3175097