University of Massachusetts Amherst

Search Google Appliance


Parente Directs Imaging Analyses for NASA’s Martian Orbiter and Rover

Mario Parente

Mario Parente

Mario Parente, an expert in the analysis of hyperspectral images and a professor in the Electrical and Computer Engineering Department, plays a critical role in a new $1.2-million National Science Foundation grant to apply recent advances in biologically inspired deep learning methods to analyze large amounts of scientific data from Mars. Parente’s research on hyperspectral camera images is being applied to direct the analysis of data gathered by a NASA orbiter, which is currently examining the chemical composition of rocks and dust on Mars.

Unlike traditional cameras, a hyperspectral instrument divides the light spectrum into many more bands than are visible to humans, according to an article posted by the UMass Amherst News Office. Such hyperspectral images have been sent back to Earth by NASA’s Mars Reconnaissance Orbiter (MRO) that has been orbiting Mars since 2003.

As another example of hyperspectral imaging, a hyperspectral map of an Amazon rainforest might identify hundreds of different tree species by analyzing images over many frequencies. The UMass News Office explains that the hyperspectral cameras aboard the Mars Recoinassance Orbiter can observe in both the visible range and shorter wavelengths, over the full infrared wavelength range. The ability to detect light in these ranges allows scientists to identify a broad range of minerals on the Martian surface.

The lead principal investigator for the NSF project is UMass computer scientist Sridhar Mahadevan from UMass Amherst’s College of Information and Computer Sciences. His co-investigators are Parente and Darby Dyar of Mount Holyoke, a specialist in planetary chemistry and geology who serves on the scientific mission team for the Mars Science Laboratory rover.

The team’s NSF proposal stated that “The central goal of the NSF research is to develop a broad innovative deep learning framework for knowledge discovery from large-scale scientific datasets, focusing on spectroscopic data from the CRISM spectrometer on MRO and the LIBS spectrometer on the Mars Science Laboratory (MSL) rover. Deep learning is an emerging area of machine learning that has been extensively studied in areas such as computer vision, speech recognition, and natural language processing. However, its efficacy at extracting structure from data produced by scientific instruments is much less understood, and yet if it were equally successful, it would surely make a significant impact in scientific data analyses.”

As part of that NSF proposal, the UMass team is conducting a thorough investigation of the applicability of new and innovative approaches in deep learning to analyze data from the Mars Reconnaissance Orbiter.

As the NSF proposal explained, “CRISM is an imaging spectrometer (or hyperspectral imager, HSI) with a field of view that can cover wavelengths from 0.362 to 3.92 microns (362 to 3920 nanometers) at 6.55 nanometers/channel. This means that CRISM can observe in both the visible range and shorter wavelengths within the infrared wavelength range. Being able to detect light in these wavelength ranges enables the CRISM team to identify a broad range of minerals on the Martian surface.”

As the NSF proposal expounded, hyperspectral images are usually acquired from an airplane or a satellite, and the instantaneous field of view (IFOV), or the area covered, by a given pixel is relatively large. At these scales, the spectrum in a pixel of a hyperspectral image is a complex combination or mixture of the light scattered by the pure components within it, called the endmembers.

Parente’s research team uses “spectral unmixing” to identify and quantify the endmembers present in each pixel of the image. As the NSF proposal noted, algorithms for spectral unmixing depend on the type of mixing present in the IFOV. Linear mixing holds when incident light rays interact with just one material before being collected by the sensor, as in checkerboard-type scenes, and the resultant mixed spectrum can be modeled as a linear combination of the pure materials in the IFOV. Conversely, nonlinear mixing results from the scattering of the light rays by multiple materials in the IFOV. These interactions can be between macroscopic objects, such as boulders and the floor of a crater, or at a microscopic, or intimate, level.

“Microscopic nonlinear mixing is prevalent on the surface of Mars and it occurs for particulate media such as mineral mixtures and soils, where grains of minerals are in close contact,” as the NSF proposal explained. “If each constituent phase has different physical and optical characteristics, incoming light will show complex nonlinear interactions between the individual mineral grains.”

Parente earned his B.S. and M.S. degrees in telecommunications engineering from the University of Federico II of Naples, Italy, and he earned M.S. and Ph.D. degrees in Electrical Engineering from Stanford University. Prior to joining UMass Amherst, he was a post-doctoral research associate at Brown University. Dr. Parente's research interests include hyperspectral imaging, sensors, and data analysis, and machine learning techniques for determining ground composition of terrestrial and planetary surfaces. Besides working on the NASA Compact Reconnaisance Imaging Spectrometer for Mars, he is supporting the NASA BASALT Science team and the MIT Lincoln Laboratory. (June 2016)