| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 2.59 MB | Adobe PDF |
Advisor(s)
Abstract(s)
O oceano é atualmente explorado a vários níveis. A nível político e económico, recorde-se que 90% do comércio mundial é feito por mar. A nível de investigação, muitos são os que se dedicam ao estudo das características do leito do mar e do próprio meio subaquático, bem como da fauna e da flora marinhas. Deste crescente interesse pelo mundo submarino, surge a necessidade de o compreender enquanto meio de transmissão e receção de sinais e a melhor forma de o fazer é através da acústica submarina, por ser a
onda do tipo sonora a que se propaga mais facilmente na água. A presente dissertação recorre a várias técnicas de data mining com o objetivo final de identificar uma fonte sonora, mais particularmente um navio, partindo de registos acústicos de navios já conhecidos. É realizado o enquadramento teórico e são apresentados os conceitos considerados pertinentes na área da acústica submarina e processamento de sinal, bem como descritos os métodos de classificação utilizados para identificar um novo registo
acústico de navio. Os dados utilizados para treinar e testar os classificadores foram fornecidos pelo Instituto Hidrográfico, tendo sido todos obtidos no exercício naval Recognized Environmental Picture em 2013 (REP 13). Na fase de testes, após o préprocessamento dos dados, os resultados apontam, de uma forma geral,
para uma elevada taxa de sucesso dos classificadores na identificação de um novo dado. Deve-se no entanto considerar uma limitação ao estudo, o facto dos registos acústicos terem sido recolhidos todos
no mesmo local, num pequeno intervalo temporal. Futuramente fará sentido aplicar este método a uma
escala maior, para que o sinal acústico de um navio por classificar seja facilmente associado ao navio a
que pertence desde que conste numa base de dados. Esta ferramenta poderá ser utilizada pela Marinha com diversos propósito, entre os quais monotorização costeira e fiscalização.
Today, oceans are persistently explored at different levels. On a political and economic level, around 90% of the world trade is conducted through sea. On a scientific level, both aquatic fauna and flora, as well as the characteristics of the seabed and the aquatic medium itself are all important objects of research. Being so, it is important to understand the sea as a mean of transmission and reception of signals, and the best way to do so is through submarine acoustics, as sound wave is the one that best propagates underwater. This thesis uses several data mining techniques to ultimately identify a determined sound source (more particularly a vessel), through the acoustic recordings of already known vessels. We present the theoric fundamentals in submarine acoustics and signal processing considered pertinent and describe the classification methods used to identify a new acoustic recording of a vessel. The data used to train and test the classifiers was provided by the Hydrographic Institute and recorded during the Recognized Environmental Picture naval exercise, in 2013 (REP 13). After preprocessing the data gathered in the test phase, the results indicate a high success rate in the classification of new data. However, it should be considered as a limitation, the fact that all the acoustic records have been collected in the same place, in a small period of time. In the future, it is important to implement this method on a larger scale, in order to easily associate the acoustic recording of an unidentified ship to it’s recording previously registered in a vessel acoustic database. This tool can be useful to the Navy for several purposes, including coastal monitoring and surveillance.
Today, oceans are persistently explored at different levels. On a political and economic level, around 90% of the world trade is conducted through sea. On a scientific level, both aquatic fauna and flora, as well as the characteristics of the seabed and the aquatic medium itself are all important objects of research. Being so, it is important to understand the sea as a mean of transmission and reception of signals, and the best way to do so is through submarine acoustics, as sound wave is the one that best propagates underwater. This thesis uses several data mining techniques to ultimately identify a determined sound source (more particularly a vessel), through the acoustic recordings of already known vessels. We present the theoric fundamentals in submarine acoustics and signal processing considered pertinent and describe the classification methods used to identify a new acoustic recording of a vessel. The data used to train and test the classifiers was provided by the Hydrographic Institute and recorded during the Recognized Environmental Picture naval exercise, in 2013 (REP 13). After preprocessing the data gathered in the test phase, the results indicate a high success rate in the classification of new data. However, it should be considered as a limitation, the fact that all the acoustic records have been collected in the same place, in a small period of time. In the future, it is important to implement this method on a larger scale, in order to easily associate the acoustic recording of an unidentified ship to it’s recording previously registered in a vessel acoustic database. This tool can be useful to the Navy for several purposes, including coastal monitoring and surveillance.
Description
Keywords
Acústica submarina, data mining, pré-processamento, classificadores, monotorização.