HSI and MSI Atmospheric Correction Using Neural Networks

"Developing real-time, unsupervised, on-board data processing algorithms for emerging remote sensing technologies is a key step towards overcoming bottlenecks in both ground-based processing and transmission capability to a ground receiving station. Animportant initial processing step is the application of an atmospheric correction algorithm (ACA), in which the effects of the intervening atmosphere are removed from hyperspectral and multispectral images (HSI and MSI). Spectral Sciences, Inc. proposesto develop a novel Neural Network (NN) based ACA for HSI and MSI sensors that can be embedded in an application specific integrated circuit (ASIC) to perform autonomous, real-time, and on-board atmospheric correction. While sophisticated, non-real timeground-based ACA's have been developed, their representation in terms of a NN has yet to be demonstrated. Significantly, the NN approach may exceed their performance, particularly for the difficult problem of aerosol characterization (visibility and typeover various surfaces). The objectives of Phase I are to demonstrate that NN's can accurately perform the ACA functions, which are atmospheric parameter retrieval and spectral reflectance calculation. In Phase II, the NN algorithms will be implemented ina complete, fully automated software package, in preparation for Phase III transitioning onto ASIC or FPGA hardware. Applications include surface terrain mapping and reflectance characte