Imaging spectrometers yield rich and informative data products, but interpreting them demands time and expertise. There is a continual need for new algorithms and methods for rapid firstdraft analyses to assist analysts during instrument operations. Intelligent data analyses can summarize scenes to draft geologic maps, searching images to direct operator attention to key features. This validates data quality while facilitating rapid tactical decision-making to select follow-up targets. Ideally, these algorithms would operate in seconds, never grow bored, and be free from observation bias about the kinds of mineralogy that will be found.
A method was developed for operatordirected search within imaging spectrometer data. Analysts can use a pure Web interface to build a library using spectra from libraries or previous images, and then search for similar spectra within catalog images. The service is analogous to a search engine capable of finding a match to a user query. The results appear in the form of a map identifying spectral signature matches to the target. Searching within a single image reveals areas of similar reflectance to assist understanding of scene composition. Searching across images can help discover mineral anomalies as well as understand trends in composition on planetary scales.
Just as Web search involves more than matching phrases, spectral search is more challenging than simply matching the shape of the spectrum. The background substrate and illumination changes can cause the same signature to appear very different across scenes. Here, the search service uses an adaptive matched-filter approach that compensates for the context and background characteristics of each scene. It searches millions of candidate pixels in seconds or less. An additional benefit is that the inverse covariance matrix (a natural byproduct of the signature matching) can be used for anomaly detection with the classic Reed-Xiaoli (RX) algorithm. The covariance is inverted using dominant mode rejection and diagonal loading as in previous work.
The target detection is not particularly novel, but it is used in a new application. At the time of this reporting, this is believed to be the first instance of a distributed Web-based system for userdefined spectral library searches in hyperspectral images. Currently, the full images are transferred manually by disk or network transfer for manipulation by specialized software that resides on the user’s computer.
The Web interface is a convenient method for exploring and browsing these data without the license requirements of full commercial hyperspectral analysis software or the challenge of distributing many Gigabyte-scale data cubes to multiple members of a distributed instrument team. It could be useful for real-time operations of imaging spectrometers during rover traverses and deep space exploration. Terrestrial crossover applications are also likely since this provides an easy way to distribute imaging spectrometer data to clients. Airborne and future spaceborne instruments will generate extremely large data catalogues, and cursory analyses will require remote, distributed tools for manipulating them.
This work was done by David R. Thompson, Alexander Smith, and Rebecca Castano of Caltech for NASA’s Jet Propulsion Laboratory.