SBIR-STTR Award

A need exists for providing timely damage assessment information to the key decision makers on a typical airbase during an attack or natural disaster. At present, the difficulty of provising accurate and timely information hinders decision making during a crisis. Real-time damage assessment may be possible through the application of neural network technology. For a damage assessment system to be successful it must perform three key functions: recognition and identification, categorization, and location. In this proposal a system consisting of a digital computer hosting a neural network with simulated sensors will be used to investigate the capabilities of the system to perform the three functions. The main thrust of this effort is to show that these functions can be accomplished on real-time and with sufficient precision and reliability to make remote sensing damage assessment possible.

Previous efforts have demonstrated that neural networks can form the basis of an automated damage assessment system which can detect, locate, and categorize explosive detonations. However, the network configurations currently in use for damage assessment do not provide the accuracy required in a deployable system. Further, the current configurations suffer considerable performance degradation in both speed and accuracy when multiple detonations occur in rapid succession. This proposal suggests research which will offer significant performance gains in both speed and accuracy for neural networks in damage assessment.

Keywords:
DAMAGE ASSESSMENTS NEURAL NETWORKS PATTERN RECOGNI REAL TIME DECISION MAKING DIGITAL COMPUTE SENSORS