Underwater target classification using canonical correlations

Undersea Target Classification Using Canonical Correlation Analysis

“…That is (4) where and are orthogonal matrices, i.e., , , and is a diagonal singular value matrix, with and . Then, the transformation (5) resolves into their canonical coordinates , with the composite covariance matrix (6) We refer to the elements of and as the canonical coordinates of and , respectively. The diagonal cross-correlation matrix (7) is called the canonical correlation matrix of canonical correlations , with…”

“…In recent years, several multiaspect-based methods for detection and classification of underwater targets from acoustic backscattered signals have been developed, e.g., see [1]- [6]. A detailed review of these methods is provided in [1].…”

A Multichannel Canonical Correlation Analysis Feature Extraction with Application to Buried Underwater Target Classification

“…This pattern of coherence is different than that found among sonar returns for non-mine-like objects. This has been demonstrated experimentally using two sonar returns [3], [8]. The multichannel extension of CCA [5] is used here for extracting features from the BOSS data set.…”

“…In the past, many different methods [2], [3] have been developed and applied to the task of correctly detecting and differentiating between mine-like and non-mine-like objects. From these studies, it has been determined that one way to effectively deal with this problem is to utilize multiple looks at an object to make a more informed classification decision.…”

“…In the first approach, common features among multiple sonar pings are identified and extracted for better object discrimination [3]. The motivation behind this approach is the idea that mine-like objects exhibit different patterns of coherence when compared to non-mine-like objects.…”