A carregar...
Publicação
Signal processing and automatic classification tools in the development of a new opto-electronic nose
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 33.38 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Bacterial infections are a worldwide concern due to the increasing microbial resistance
to antibiotics. Therefore, a need to create fast diagnose methods has risen.
Electronic noses are devices that try to mimic the olfactory system. These systems
became popular due to their fast response time and portability, and for that reason, they
are seen as a possible diagnose method.
In the Biomolecular Engineering laboratory, a project involving an electronic nose is
being developed, in which the final goal is the diagnosis of bacterial infections.
The objective of the present dissertation was to develop an analysis tools to complement
the system that is being developed.
First, some preprocessing methods were chosen and applied to the acquired data, then
a classification tool was developed. Machine learning algorithms were used: a recursive
feature selection method was applied to select the best features to characterize the signals
and a Support Vector Machine classifier trained to distinguish eleven volatile classes.
Five experiments were analysed and three different sensor formulations tested. Since
the device is yet not fully developed, samples which were used were not from bacteria.
Instead, simple volatile organic compounds were used.
The results showed that it was possible to efficiently distinguish all compounds with
the proposed methods. Moreover, important conclusions related with the current state
of the project where drawn. Namely, sensor stability is possible during long, continuous
periods of time, but limitations in the reproducibility of the production method may
influence the performance of the classifier.
Descrição
Palavras-chave
Electronic nose volatile organic compounds machine learning recursive feature elimination Support Vector Machine