Model-Based Object Identification & Multisensor Fusion of Images of Satellites

An approach is presented to solve both the data fusioin problem, and the space object model building requirement for the Air Force Phillips Laboratory. Artificial neural network (ANN) technology provides a number of tools that form the basis of this proposal and the foundtion for the solution to the real-time data fusion problem. ANNs are a proven technology that offer a powerful, naturally parallel computational technique for highly distributed processing. The need is for procedures that can readily adapt to additional target classes, new sensors and widely variable viewing environments. Finally, effective target identification and characteization is greatly enhanced if a priori knowledge about the target is integrated into the data processing. A model-based process is presented to update and enhance the spectral properties, thermal condition, orientation, and the spacecraft's geometric configuration. A model-based fusion process based on feature extraction using a neocognitron neural network will be developed to integrate spatial and spectral data nad create or update a three-dimensional satellite model. The fusion process will specifically accommodate wide band radar (Inverse Synthetic Aperture Radar-ISAR), visible, and infrared wavelength imagery and signature information of low Earth orbiting satellites. The final high fidelity satellite model will be protable through the International Graphics Exchange Standard (IGES).