Abstract:

Automated anomaly and target detection are commonly used as a prescreening step within a larger target detection and target classification framework to find regions of interest for further analysis. Many anomaly and target detection algorithms in the literature have been developed for application to target detection in Synthetic Aperture Sonar (SAS) imagery which produces high quality intensity images of underwater scenes. In this paper, a comparison of unsupervised and supervised anomaly and target detection algorithms for target detection in synthetic aperture sonar is presented.

Three different experiments are done on two subsets of a large SAS dataset containing images captured in high frequency and low frequency. Template matching and statistical methods are used in order to detect targets within the SAS imagery. For detection, it is desired that a low number of false alarms be detected with a high number of true positives. This means that a detector is able to distinguish a target from an anomaly well. In practice, it is likely that human intervention is required to confirm targets. If a detector returns a low number of false alarms the amount of labor necessary by an operator is minimal.

The purpose of the first two experiments is to observe the effects of varying certain parameters for detection. Among these parameters are the number of templates, dimensions and frequency of the SAS images and methods used to process the images. For the template matching methods, K-Means and K-Medoids are employed to generate templates of the targets to be detected. Six different detectors are utilized: adaptive cosine estimator, squared adaptive cosine estimator, spectral matched filter, matched filter, normalized cross-correlation, and convolution. Two statistical detectors are utilized: Reed-Xiaoli and Reed-Xiaoli Left-Right. The results of this thesis show that template-based detection produces comparable results to statistics-based detectors. The spectral matched filter and Reed-Xiaoli detectors show the most promising results for anomaly and target detection when low frequency data is used.

Links:

 

Citation:

P. Lyons, “Anomaly and Target Detection in Synthetic Aperture Sonar,” M.Sc. Thesis, Gainesville, FL, 2019. 

@MastersThesis{Lyons2019AnomalyDetectionThesis,
author = {Princess Lyons},
title = {Anomaly and Target Detection in Synthetic Aperture Sonar},
school = {Univ. of Florida},
year = {2019},
address = {Gainesville, FL},
month = {October},
}