Fusion of SAR and Hyper-Spectral Sensor Data for Improved ATC

The overall goal of the proposed effort is to develop a fusion process that integrates SAR and hyperspectral sensor data in order to overcome the current performance limitations of single sensor automatic target cueing. In the Phase I effort, we will define a concept of operation for SAR/hyperspectral automatic target cueing. The nature of the imaging domains of SAR and HS sensor as well as the concept of operation drive the choice of the best data fusion method. Fusion between SAR and HS imagery presents a problem in that they are orthogonal rather than coincident, and that the image information is highly uncorrelated. We will develop and demonstrate the potential of a practical fusion technique that symbolically or synergistically combines features from each sensors on an object basis rather than on a pixel basis. In particular, we will investigate a hybrid fusion method that utilizes the advantages from both centralized and distributed fusion techniques. We will demonstrate the feasibility and potential performance gains of our fusion approach using selected imagery. A major component that will be addressed during the Phase I program is correlating target regions identified in the imaging domain of the second sensor. We will estimate the expected accuracy of the correlation method assuming that accurate GPS and INS information is available. We will utilize our current stand alone automatic target cueing algorithms, based on geometric whitening filters, for SAR and multispectral imagery. These filters are implemented with grayscale morphological operations (fixed kernels for SAR and adaptive kernels for MS/HS sensors). The discrimination stage of the cueing process utilizes fractal-based texture measures that are implemented in a computationally efficient form.