Please use this identifier to cite or link to this item:
|Title:||Methods for the extraction and classification of transient signals from noisy data - A case study in classifying sounds from the thorax.|
|Presented at:||University of Leicester|
|Abstract:||The physiological origins and physical characteristics of sounds from the thorax have been reviewed briefly. This thesis presents some signal processing algorithms and classification techniques which have been developed for the extraction and classification of those sounds. In order to evaluate the recording equipment and signal processing algorithms two simulators were constructed: a laboratory simulator generating lung sounds in a variable background noise environment and a heart sound simulator written such that the generated sound was defined by a set of variables. Four conventional transformation algorithms for the transient extraction process were evaluated. Their considerable user intervention and inconsistent transformed signal led to the development of the "signal's envelope" algorithm. The signal's envelope method was used to extract transients of interest which were then used for the classification stage. It is shown that, due to the numerical nature of the features used for the classification process, the nearest neighbour clustering algorithm could not correctly classify all the extracted transients. The numerical features were therefore converted into linguistic terms and a fuzzy logic technique was developed to classify the transients. The fuzzy inference engine was robust enough to cope with the small numerical variation in features such that the correct classification was achieved. The other classification method tried was the fuzzy "min-max" clustering algorithm. This also used numerical features for the classification process and was therefore not able to classify all of the extracted transients correctly. A lung sound analyser was constructed using the signal's envelope and fuzzy inference engine. The system was able to extract and classify individual heart sounds, crackles and wheezes from recorded phonograms. In about 4% of cases, the heart sounds were so indistinct that only a partial classification was achieved. It was concluded that by using simple transducers and sophisticated signal processing and classification algorithms it was possible to construct a chest sound classifier which may be of use in a clinical environment.|
|Rights:||Copyright © the author. All rights reserved.|
|Appears in Collections:||Theses, Dept. of Engineering|
Items in LRA are protected by copyright, with all rights reserved, unless otherwise indicated.