Hyperspectral Resolution Enhancement with Support Vector Machines

With the continued growth of hyperspectral sensor technology for both military and commercial applications, there is an ever-increasing need to make the maximum use of data from these devices. Despite the wealth of information that can be extracted fromhyperspectral cubes, additional information could be extracted if these devices had better spatial resolution. The resolution of these devices is inherently limited by physical constraints (cost, size, and manufacturing). To overcome this limitation,Photon Research Associates (PRA) proposes a unique signal processing approach which combines the proven mathematics of wavelet denoising along with Support Vector Machines (SVM) for resolution enhancement of the hyperspectral data. Support Vector Machinesare based on the field of mathematics called Statistical Learning Theory develop by Vapnik at Bell Labs. SVMs have been proven to be superior to classical techniques and have been applied successfully to various application areas including patternrecognition, regression, and estimation for real valued data. We propose to extend the theory and apply SVMs for estimating and extracting the higher spatial frequencies and hence providing resolution enhancement of the hyperspectral data.The results ofthis SBIR can be applied commercial and military hyperspectral remote sensing systems in order to improve its resolution beyond the diffraction limit. Potential military applications include target/anomaly detection, site monitoring, and trafficabilityanalysis. Potential commercial applications include remote sensing for geological, land use monitoring, agricultural, environmental monitoring, and mineral exploration purposes.