Detecção de novidade para sistemas de sonar passivo

Abstract

Sound is a mechanical wave that propagates over great distances in the oceans and it can, therefore, be used for vessel detection and classification in underwater environments, which are basic sonar system tasks. The development of such systems is directly linked to the country defense, especially, in countries with continental dimensions, such as Brazil. Recently, the Brazilian Navy defined underwater acoustics as a strategic priority area. Passive sonar systems can be installed to monitor the Brazilian coast in a stealthy and efficient way. In addition, these are used in military submarines for different applications. As in this operating environment, each ship has a unique acoustic signature, and ships whose data have not been acquired can be observed, it is necessary to develop a novelty detector operating in conjunction with the contact classifiers implemented in Brazilian Navy systems. Because classification systems operate competing for computing resources with novelty detectors, they can impact in classification efficiency. The number of classes in this environment is very large, and because of this, specific performance indices were created to evaluate the developed model efficiency. In addition, different data compressors were developed to access relevant ship information of, among them can be cited PCD, kPCA, NLPCA and SAE. The novelty detection development was based on the operating environment of the Brazilian Navy and since it can have its operating conditions changed over time, a stationarity monitoring system based on higher order statistics was proposed. Both the novelty detector and the stationarity monitoring system were developed with experimental data provided by the Brazilian Navy.