|
|
|
A neural map view of planetary spectral images for precision data mining and rapid resource identifi
|
|
|
| This project follows up on a current three-year AISRP grant, NAG5-10432, which will end 8/2004. It addresses a pressing need for rapid yet intelligent analysis of voluminous multi- and hyperspectral images in order to extract key data and generate knowledge. Spectral imaging plays a leading role in remote identification of surface materials of Earth (Landsat, AVIRIS, Hyperion), Mars (Pathfinder, MGS, MER), the Jovian system (Galileo NIMS), the Saturnian system (Cassini VIMS) and other solar system bodies. Hyperspectral sensors, in particular, enable detailed identification through the complexity of signatures measured in hundreds of narrow spectral bands. The challenge in automated and fast (real time, on-board) interpretation of these huge images calls for massively par-allel algorithms, as well as it requires sophisticated algorithms for optimal knowledge extraction. Properly utilized, Artificial Neural Networks (ANNs) can provide both. The current project engineered ANNs, specifically Self-Organizing Maps (SOMs) and their hybrids for efficient and sophisticated clustering and classification of spectral images, developing custom modules supported by commercially available components. Based on some of the latest theoretical research on SOMs, the tools we developed, jointly with European experts, are powerful for distinguishing a large number of spectral classes and for the discovery of ''interesting'' but uncommon and spatially very small classes. We use information theoretically principled SOM approaches, which increases power and confidence in autonomous data mining. We demonstrated the effectiveness and high quality of data analysis on sample IMP spectral images, Cassini VIMS Jupiter fly-by im-agery, AVIRIS and other data representing typical challenges in NASAís missions. We propose to advance these computational intelligence capabilities in three ways: 1) We will add significantly new theoretical strength to information extraction modules. 2) The software, HYPEREYE, will be made transferable to other users through high-level graphic interface, augmented software design, tutorials and wrapping, opening an important phase of technology infusion that will take recent and future developments into the user community. 3) We will directly participate, using our methods and software, in analyses of spectral images forthcoming from the Mars Exploration Rovers and Cas-sini VIMS Saturn orbital tour, and (pending its funding) a Pluto/icy satellites spectral analysis project. The ëneuralí core of our software is already suitable for implementation in high-speed massively parallel hardware (which could be an on-board analysis capability), as it was one of the original objectives of our work. We are pursuing that line of development outside of this project proposal and, if successful, we anticipate using the hardware to support this work as well. |
Bibliography
| | Relevance-based Feature Extraction for Hyperspectral Images
| |
| Advances in computational intelligence and learning
Michael Biehl, Erzsébet Merényi and Fabrice Rossi, Neurocomputing 70(7-9), editorial. pp 1117-1119
| |
| Min(d)ing the small details: discovery of critical knowledge through precision manifold learning and application to on-board decision support
Merényi, E., L. Zhang, and K. Taşdemir, Proc. IEEE Int�l Conference on Systems of Systems Engineering (IEEE SoSE 2007), San Antonio, TX, April 16 � 18, 2007. 8 pp
| |
| Data topology visualization for the Self-Organizing Map
Taşdemir, K. and Merényi, E., Proc. 14th European Symposium on Artificial Neural Networks, ESANN'2006, Bruges, Belgium, 26-28 April, 2006. pp. 125-130
| |
| Spectral Class Distinctions Observed in the MPF IMP SuperPan Using a Self-Organizing Map.
Farrand, William, H.; Erzsebet Merenyi; Scott Murchie; Oliver Barnouin-Jha, Proc. 36th Lunar and Planetary Science Conference, Houston, Texas, March, 2005., (March, 2005) pp. 2 pp.
| |
| Intelligent Understanding of Hyperspectral Images through Self-Organizing Neural Maps
Merenyi, Erzsebet; Proc. 2nd Int'l Conf. Cybernetics and Information Technologies, Systems and Applications (CITSA 2005), July 14 - 17, 2005, Orlando, FL, USA, (July, 2005) N/A pp. 30 - 35
| |
| Neural Maps for Precision Data Mining: Application to Planetary Spectral Images
Merenyi, Erzsebet; Proc. Jordan Int'l Conference in Computer Science and Engineering, Al-Balqa University, Salt / Amman, Jordan, Oct 4 -7, 2004, (Oct, 2004) N/A pp. 10 pp.
| |
| The Use of AVIRIS Imagery to Assess Clay Mineralogy and Debris-Flow Potential in Cataract Canyon, Utah
Rudd, Larry; Erzsebet Merenyi; Geological Society of America Abstracts with Programs, (November, 2004) 36(5) pp. 385 - 385
| |
|
|
