Mark Dzwonczyk
Sensor and Commun. Electron. Div., C. S. Draper Lab., 555 Technology Square, Cambridge, MA 02139
Computational architectures modeled after the biological nervous systems, so-called neural networks, have been shown to excel at pattern classification problems where the input data set is large and corrupted by noise and the solution formulation is not well defined. This paper discusses the application of neural computing techniques to a target detection problem with side-scan sonar data. A brief review of the fundamental concepts of feed-forward neural networks is first presented. Image segmentation and classification with a large analog electronic neural network are then described. It is shown that such neural computing approaches can be expected to be more robust than conventional statistical classifiers and are far better suited for deployed, real-time implementation of the system architecture.