Comparison of Different Classification Algorithms for Underwater Target Discrimination

A Review of Underwater Mine Detection and Classification in Sonar Imagery

Underwater mines pose extreme danger for ships and submarines. Therefore, navies around the world use mine countermeasure (MCM) units to protect against them. One of the measures used by MCM units is mine hunting, which requires searching for all the mines in a suspicious area. It is generally divided into four stages: detection, classification, identification and disposal. The detection and classification steps are usually performed using a sonar mounted on a ship’s hull or on an underwater vehicle. After retrieving the sonar data, military personnel scan the seabed images to detect targets and classify them as mine-like objects (MLOs) or benign objects. To reduce the technical operator’s workload and decrease post-mission analysis time, computer-aided detection (CAD), computer-aided classification (CAC) and automated target recognition (ATR) algorithms have been introduced. This paper reviews mine detection and classification techniques used in the aforementioned systems. The author considered current and previous generation methods starting with classical image processing, and then machine learning followed by deep learning. This review can facilitate future research to introduce improved mine detection and classification algorithms.

Geometry and topology of the space of sonar target echos

Successful synthetic aperture sonar target classification depends on the "shape" of the scatterers within a target signature. This article presents a workflow that computes a target-to-target distance from persistence diagrams, since the "shape" of a signature informs its persistence diagram in a structure-preserving way. The target-to-target distances derived from persistence diagrams compare favorably against those derived from spectral features and have the advantage of being substantially more compact. While spectral features produce clusters associated to each target type that are reasonably dense and well formed, the clusters are not well-separated from one another. In rather dramatic contrast, a distance derived from persistence diagrams results in highly separated clusters at the expense of some misclassification of outliers.

SETTING UP A PROBABILISTIC NEURAL NETWORK FOR CLASSIFICATION OF HIGHWAY VEHICLES

Many neural network methods used for efficient classification of populations work only when the population is globally separable. In situ classification of highway vehicles is one of the problems with globally nonseparable populations. This paper presents a systematic procedure for setting up a probabilistic neural network that can classify the globally nonseparable population of highway vehicles. The method is based on a simple concept that any set of classifiable data can be broken down to subclasses of locally separable data. Hence, if these locally separable data can be identified, then the classification problem can be carried out in two hierarchical steps; step one classifies the data according to the local subclasses, and step two classifies the local subclasses into the global classes. The proposed approach was tested on the problem of classifying highway vehicles according to the US Federal Highway Administration standard, which is normally handled by decision tree methods that use vehicle axle information and a set of IF-THEN rules. By using a sample of 3326 vehicles, the proposed method showed improved classification results with an overall misclassification rate of only 2.9% compared to 9.7% of the decision tree methods. A similar setup can be used with different neural networks such as recurrent neural networks, but they were not tested in this study especially since the focus was for in situ applications where a high learning rate is desired.

Multiple angle acoustic classification of zooplankton

The use of multiple angle acoustic scatter to discriminate between two taxa of fluid-like zooplankton, copepods and euphausiids, is explored. Using computer modeling, feature extraction, and subsequent classification, the accuracy in discriminating between the two taxa is characterized via computer simulations. The model applies the distorted wave Born approximation together with a simple system geometry, a linear array, to predict a set of noisy training and test data. Three feature spaces are designed, exploiting the relationship between the shape of the scatterer and angularly varying scattering amplitude, to extract discriminant features from these data. Under the assumption of uniform random length and uniform three-dimensional orientation distributions for each class of scatterers, the performance of several classification algorithms is evaluated. Simulations reveal that the incorporation of multiple angle data leads to a marked improvement in classification performance over single angle methods. The improvement is more substantial using broadband scatter. The simulations indicate that under the stated assumptions, a low classification error can be obtained. The use of multiple angle scatter therefore holds promise to substantially improve the in situ acoustic classification of fluid-like zooplankton using simple observation geometries.