| Summary: | 博士 === 國立臺灣大學 === 工程科學及海洋工程學研究所 === 107 === Underwater acoustic signature identification has been employed as a technique for detecting underwater vehicles, such as in anti-submarine warfare or harbor security systems. The underwater sound channel, however, has interference due to spatial variations in topography or sea state conditions and temporal variations in water column properties, which causes multipath and scattering in acoustic propagation. Thus, acoustic data quality control can be very challenging. One of challenges for an identification system is how to recognize the same target signature from measurements following the rule of DNV GL-Silent under different temporal and spatial settings.
This paper deals with the above challenges by establishing an identification system composed of feature extraction, feature selection, and data mining with statistical analysis and machine learning approaches to recognize the target signatures of underwater radiated noise from a research vessel, Ocean Researcher III (OR3), with a bottom mounted hydrophone in four cruises in 2016 and 2017. The fundamental frequency and its power spectral density are significant features for classification. In feature extraction, we extract the features before deciding which of the two aforementioned features is more significant. The first approach utilizes Polynomial Regression (PR) classifiers and feature selection by Taguchi method and analysis of variance (ANOVA) for finding the maximum desirability function under a different combination of factors and levels. The second approach utilizes Adaptive Network-Based Fuzzy Inference System (ANFIS) and selecting the optimized desirability value via genetic algorithm (GA).
The result proves that the feature extraction process successfully detects the OR3 targets under different topography and temporal variations and solves the noise masking problem by utilizing harmonic frequency features extracted from unmasking the frequency bandwidth for ship noises. Nevertheless, the accuracy of PR and ANFIS classifiers all approach 90%. The “Artificial Intelligence Application in a Marine Vehicle Identification System Using Acoustic Signatures” developed here can be carried out in harbor security for monitoring acoustic signatures of a specific ship. In addition, it utilizes the artificial intelligence classifier, which has faster detecting performance than the regression classifier on real-time monitoring.
|
|---|
