Rapid Identification of Planetary Resources from Spacecraft
Data transmission severely limits the return of raw data from high-rate instruments on deep space missions. For example, a high-resolution imaging spectrometer at Mars could collect data more than 106 times faster than the best available downlink rate. There is a pressing need for rapid yet intelligent on-board analysis for selection of key data to return to Earth (along with generated knowledge). The highest priority observations, such as compositional mapping over hydrothermal vents, may require only a small fraction of the total data volume.
We will apply state-of-the-art parallel computational technology, a hybrid neural network with a Self-Organizing component, for fast clustering and classification of planetary multispectral data. Our approach is powerful because: (1) the massively parallel algorithm can be readily implemented in hardware to process a high-volume data stream in (near) real time, and (2) neural network clustering and classification is superior to conventional methods. This approach will enable space scientists to extract key information from high-rate imaging spectrometers or other instruments expected from NASA's flight missions. We will develop automated processes to evaluate and render cluster information, integrated with pre-processing and supervised classification, to be used for on-board analysis with limited human interaction as well as for the analysis of large data sets on the ground. We will build on existing and commercially available products as well as on our previous work, and test the resulting tools on Clementine and Mars Pathfinder multispectral data sets and AVIRIS hyperspectral image cubes. The product will be applicable to a wide range of planetary missions and will be made available to the scientific community. We plan to follow up with a separate project to implement the software into a high speed parallel hardware board specifically tailored for an instrument that is being planned at LPL for a Mars orbiter mission.
This project is a collaboration between computer science and space science investigators within LPL, University of Arizona.
Back to Artificial Neural Network Page