An Innovative Approach for the Development of Common Terrain Databases

The derivation of spatially correlated high-fidelity Infrared (IR) and Radar (SAR) scenes from visual imagery texture maps used in visual simulators requires: 1.) the accurate, timely creation of Wavelength Independent Texture (WIT) map databases and 2.) a simulator capable of using these databases to produce simulated IR/Radar imagery. The primary obstacle to widespread use of IR/Radar with real-time out-the-window simulations is the low level of automation in the development of WIT maps. This Phase I effort will define, prototype and evaluate a system for the rapid production of WIT maps from visible imagery and numerous disparate data layers, including raw and processed NIMA, DTED and DFAD data and digital cartographic data. Neural networks will be trained on imagery to recognize the spectral and collateral inputs. The use of collateral inputs offers a vast improvment over previous neural networks material classification research which focuses solely on spectral characteristics of the data. The system will also accommodate the use of single-band imagery. In this effort PRA will prototype and evaluate a mapping system which is expected to improve the accuracy of WIT maps and reduce the develop-ment time for WIT maps from weeks to hours.