Undersea Target Classification Using Canonical Correlation Analysis

Abstract

Abstract-Canonical correlation analysis is employed as a multiaspect feature extraction method for underwater target classification. The method exploits linear dependence or coherence between two consecutive sonar returns, at different aspect angles. This is accomplished by extracting the dominant canonical correlations between the two sonar returns and using them as features for classifying mine-like objects from nonmine-like objects. The experimental results on a wideband acoustic backscattered data set, which contains sonar returns from several mine-like and nonmine-like objects in two different environmental conditions, show the promise of canonical correlation features for mine-like versus nonmine-like discrimination. The results also reveal that in a fixed bottom condition, canonical correlation features are relatively invariant to changes in aspect angle.Index Terms-Canonical correlations, linear dependence and coherence, multiaspect feature extraction, underwater target classification.

Buried Underwater Object Classification Using a Collaborative Multiaspect Classifier

“…This is due to this property that, in the CCA, the level of coherence between the same frequency subband from two different pings can be used as a measure to detect potential mine-like objects. On the other hand, the pattern of coherence between these frequency subbands, captured by the dominant canonical correlations, would allow the discrimination of the detected objects [2]. This is accomplished by first extracting coherence-based frequency subband features for all pings in a given run through a target field and then applying them to the appropriate classification system in order to determine the location and type of the objects.…”

“…We also assume that based only on its observation , the th agent makes a single local preliminary decision using (2) where is the mapping function of the PNN classifier that captures the decision rule. To obtain a final decision , a decision rule is also used such that…”

“…The size of each subband is five samples, with each sample corresponding to a frequency bin of 10.5 Hz. This subband size was chosen to provide a small enough window for the CCA process to capture the subtle information in each frequency subband, as well as to avoid data poverty when forming the sample covariance matrices needed in the CCA [2] by averaging over the data from the 40 hydrophone channels.…”

“…One way to accomplish this is to use canonical correlation analysis (CCA) [1], which allows coherence-based features to be extracted from pairs of sonar pings. Using these coherence-based features allows for successful discrimination between mine-like and nonmine-like objects [2], [3].…”

On Measure Transformed Canonical Correlation Analysis

“…They are easily obtained by solving a simple generalized eigenvalue decomposition (GEVD) problem, which only involves the covariance and cross-covariance matrices of the considered random vectors. LCCA has been applied to blind source separation [3], image set matching [4], direction-of-arrival estimation [5], [6], data fusion and group inference in medical imaging data [7], localization of visual events associated with sound sources [8], audio-video synchronization [9], undersea target classification [10] among others.…”