Hyperspectral Image Processing

The ProblemThere has been significant recent interest in hyperspectral sensing technology to support a variety of remote sensing applications. The wealth of information that resides in the spectral domain provides significant enhancements relative to traditional panchromatic and multispectral imagery. For example, there is the potential to classify scene elements with subtle material differences as well as detect low contrast targets in complex background clutter. However, the design and development of practical hyperspectral sensors often results in a significant trade-off in spatial resolution. Therefore, important spatial features such as shape and texture can be lost, and the spatial fidelity of the resulting hyperspectral products is reduced.

This inherent trade-off between spatial and spectral quality has resulted in the development of remote sensing systems that include low resolution hyperspectral coupled with high resolution panchromatic and/or multispectral imaging subsystems. As examples, the NASA EO-1 satellite includes a 30 m hyperspectral sensor and a 10 m panchromatic imager, and the Orbview-4 satellite was designed to incorporate an 8 m hyperspectral sensor with a 1 m panchromatic imager and 2 m multispectral imager. These sensor system approaches provide an opportunity to jointly process the hyperspectral and high-resolution imagery to achieve improved detection and/or classification performance.

We have developed a novel statistical estimation framework for estimating the underlying high spatial resolution hyperspectral scene using the observed low-resolution hyperspectral data and a correlated (and co-registered) high-resolution image from another sensor. We are also studying novel anomaly detection and change detection algorithms.

Depiction of our hyperspectral image fusion process. Low spatial resolution hyperspectral imagery is fused with high-resolution broadband (grayscale) imagery to produce high spatial resolution hyperspectral imagery. Our approach seeks to fully exploit global and local correlations between the input imagery and is one of the first we are aware of to use a maximum a posteriori estimation approach to address this problem.

Selected References

P. C. Hytla, R. C. Hardie, M. Y. Eismann, and J. Meola, “Anomaly Detection in Hyperspectral Imagery: A Comparison of Methods Using Diurnal and Seasonal Data,” SPIE Journal of Applied Remote Sensing, Vol. 3, 033546 (2009); doi:10.1117/1.3236689, 2009.
M. T. Eismann, J. Meola, and R. C. Hardie, “Hyperspectral change detection in the presence of diurnal and seasonal variations,” IEEE Transactions on Geoscience and Remote Sensing, Vol. 46, pp. 237-249 (January 2008).
M. T. Eismann and R. C. Hardie, “Hyperspectral resolution enhancement using high resolution multispectral imagery with arbitrary response functions,” IEEE Transactions on Geoscience and Remote Sensing, vol. 43, pp. 455-464 (March 2005).
M. T. Eismann and R. C. Hardie, “Stochastic spectral unmixing with enhanced endmember class separation,” Applied Optics, vol. 43, pp. 6596-6608 (April 2004).
M. T. Eismann and R. C. Hardie, “Application of the stochastic mixing model to hyperspectral resolution enhancement,” IEEE Transactions on Geoscience and Remote Sensing, vol. 42 pp. 1924-1933 (September 2004).
R. C. Hardie, M. T. Eismann, and G. L. Wilson, “MAP Estimation for hyperspectral image resolution enhancement using an auxiliary sensor,” IEEE Transactions on Image Processing, vol. 13, pp. 1174-1184 (September 2004).