In this paper the automated detection and classification of mine-like objects in sidescan (Marine Sonics) data is investigated. In particular, 4 different sites are considered involving sidescan sonar data collected by Australian, New Zealand, and NATO Undersea Research Centre (NURC) REMUS AUVs. Twelve different automated detection methods are proposed and their individual and fused performance...

In this paper we investigate techniques to train an agent to accomplish certain tasks. Artificial Neural Networks will be the technique used to the train the agent. This paper will investigate the use of Generalised Regression Neural Network (GRNN) to create and train agents capable of detecting face images. This agent would make up the ‘Detection Agent’ in an architecture comprising of several...

Recent developments in target decomposition theorems indicates that the polarimetric signature of a target describes scattering mechanisms, such as depolarization, even bounce, or odd bounce, that may assist in the differentiation of a man-made targets from natural clutter, a critical first step in Automatic Target Recognition (ATR). Cloude’s alpha-entropy decomposition of the coherency matrix,...

1) Target recognition via radar cross section (RCS) profiles: In this approach, databases of the RCS of targets at different incident and observed angles are created using method-of-moments computational electromagnetics codes. The extracted RCS profiles for different targets, scaled to account for antenna patterns and atmospheric propagation, are compared to the collected data. A coordinated f...

One of the most challenging problems in automatic target recognition is reliable detection of targets in high clutter backgrounds. When the clutter statistics are unknown or highly variable, the false alarm rate of classical detection algorithms , e.g. the matched lter, cannot be controlled and target detections become unreliable. The reason for this is lack of robustness of the test statistics...

Template-based automatic target recognition (ATR) algorithms such as the Synthetic Aperture Radar Target Location and Recognition System (STARLOS) algorithm typically use separate templates to represent target signatures for ranges of articulations, aspect, depression, and squint angles. There is a performance tradeoo between ATR accuracy and the number of templates used. We use a hybrid model/...

AImtraet--The utility o f combining neural networks with pyramid representations for target detection in aerial imagery is explored. First, it is shown that a neural network constructed using relatively simple pyramid features is a more effective detector, in terms o f its sensitivity, than a network which utilizes more complex object-taned features. Next, an architecture that supports coarse-t...

This paper examines metrics for measuring clutter eeectiveness on model-based automatic target recognition systems with FLIR sensors. The measure for clutter eeectiveness proposed is the diierence of two Kullback-Leibler distances between the idealized approximate probabilistic models without clutter and the real model containing clutter. We establish that occluding objects and clutter, when ma...

Attributed scattering feature models have shown potential in aiding automatic target recognition and scene visualization from radar scattering measurements. Attributed scattering features capture physical scattering geometry, including the non-isotropic response of target scattering over wide angles, that is not discerned from traditional point scatter models. In this paper, we study the identi...

I introduce a way of verifying phonological processes on the basis of phonetic substances obtained by automatic speech recognition. The acoustic characteristics of phone-like units are modelled using automatic speech recognition techniques. This phone recogniser is run on the data tokens whose segmental structure matches with the context of target processes to be verified. Examining output stri...