Hyperspectral Imaging and its Applications

Russ Islam
October 26, 2015

Submitted as coursework for PH240, Stanford University, Fall 2015

Fig. 1: A hyperspectral cube of Moffett Field, California (Courtesy of NASA/JPL-Caltech).

Materials that appear similar under visible light may appear different under infrared or ultraviolet light. What if one were to use images under those three sets of wavelengths? Why stop at just three images? Why not use a continuum of wavelengths to collect a massive number of images? Hyperspectral imaging makes use of electromagnetic radiation at wavelengths both inside and outside of the visible spectrum to characterize materials and objects. [1] Aircraft often take hyperspectral images of the surface of the Earth for use in applications including oil spill detection and hazardous waste monitoring. Before considering these applications, a brief overview of the technical details is provided.

Technical Background

The information collected by a hyperspectral imaging system, a hyperspectral cube, is actually three-dimensional. In addition to width and height, there is a depth component associated with the wavelength at which each photograph was taken, meaning that the hyperspectral cube is simply a stack of several images. For example, a hyperspectral cube of Moffett Field, California is shown in Fig. 1, where we can see the landscape at different wavelengths by looking at the top face of the stack and gradually moving downward. The wavelength dimension usually has a resolution of 10 to 20 nanometers, so hyperspectral imaging produces very rich datasets. [1] For a single point on the surface of the Earth, hyperspectral imaging produces several reflectance values, one for each wavelength, instead of only the single optical pixel that one would obtain from a standard photograph. If we were to choose a particular spatial location (a particular pixel) and only adjusted the wavelength, we would obtain the spectrum at that location. Therefore, these spectra can be used to identify the materials located at certain pixels. Consider Fig. 2, where the spectra of a handful of minerals are extracted from a hyperspectral cube. The graph on the right-hand side shows various spectra that differ greatly among the constituent minerals.

An outline of the process of extracting useful information from a hyperspectral cube is as follows: [2]

  1. Acquisition of raw hyperspectral cube

  2. Data preprocessing, including normalization, image calibration, and correction for scattering by the atmosphere

  3. Dimensionality reduction (e.g., principal components analysis)

  4. Feature extraction

  5. Spectral signature analysis (e.g., comparison to a database of known signatures)

Fig. 2: Mineral spectra computed from hyperspectral images of stones (Source: Wikimedia Commons).

This general procedure incorporates standard image processing and machine learning techniques to deal with the high dimensionality of hyperspectral data. Because the wavelength of imaged electromagnetic radiation is adjusted in very small intervals, adjacent images in the hyperspectral cube are highly correlated. [2] For example, if one were to capture an image at the wavelength of visible red light and then were to capture a second image at the wavelength of slightly brighter red, the two images would appear very similar. This means that nearby images in the hyperspectral cube are somewhat redundant, and therefore techniques such as principal components analysis are used to mitigate this redundancy.

Though it depends on the particular application, feature extraction involves processing the dimension-reduced data in such a way as to generate higher-level information for use in classification. [2] Machine learning techniques can be used to find these features in the data. Finally, once the spectrum of a certain pixel has been processed, it is compared to the spectra of other pixels or to the spectra of known materials. A commonly used similarity measure is the angle between two spectra, where each spectrum is considered to be a vector. [3] An imaged pixel whose spectrum is sufficiently similar to that of a known material can be classified as that material.

Oil Spill Detection

Hyperspectral imaging systems aboard aircraft and spacecraft can detect hydrocarbons and played an important role in the response to the Deepwater Horizon oil spill. [4] Because even trained experts sometimes mistake marine or biological phenomena for oil spills when looking at only photographs in the visual spectrum, hyperspectral imaging is particularly useful. Many systems analyze collected data in real-time and automatically identify potential locations of oil spills and seeps. Of particular use in detecting petroleum is the near infrared portion of the electromagnetic spectrum, where carbon-hydrogen bonds produce characteristic overtones. Thermal infrared is useful as well because the emissivity of oil and seawater are different even at night. This is a result of the lower heat capacity of oil relative to water. [4] However, because oil spills evolve rapidly, hyperspectral data from aircraft and spacecraft can become obsolete in less than a day, so it is necessary for systems to collect data often and to process it quickly.

Hazardous Waste Monitoring

Hyperspectral imaging systems find a broad range of uses regarding hazardous materials management. Images of mine waste can be quickly analyzed to search for minerals that produce acids that pollute rivers. Heavy metals, such as cadmium, lead, and arsenic, are also monitored using hyperspectral imaging. Interestingly, hyperspectral data can also be used to characterize vegetation, which is then used to infer the mineral composition of the underlying soil. [5]

Thermal imaging has also been successfully used to locate burial sites of hazardous materials from weapons development projects by distinguishing between disturbed and undisturbed soil. Additionally, underground fires in mines and landfills have also been detected and monitored by hyperspectral imaging systems. On the other hand, hyperspectral imaging has also produced maps of minerals in areas such as the Rockies, which is useful for resource exploration. [5]

An Emerging Technology

Hyperspectral imaging offers clear benefits and has been instrumental in geospatial exploration and monitoring. Unlike reflection seismology, which uses controlled seismic energy sources like dynamite to produce subsurface images of the Earth, hyperspectral imaging simply uses electromagnetic radiation, which is unobtrusive but can only be used to map surfaces. [6] As more aircraft and spacecraft become equipped with hyperspectral sensors and as image processing algorithms improve, the use of hyperspectral imaging in the energy industry and for environmental applications will increase. Such advances would offer a more complete picture of the world and its resources as well as improve disaster response capabilities.

© Russ Islam. The author grants permission to copy, distribute and display this work in unaltered form, with attribution to the author, for noncommercial purposes only. All other rights, including commercial rights, are reserved to the author.


[1] P. Shippert, "Why Use Hyperspectral Imagery?" Photogramm. Eng. Rem. S. 70, No. 4, 377 (April 2004).

[2] M. Alam and P. Sidike, "Trends in Oil Spill Detection via Hyperspectral Imaging," IEEE 6571686, 7th Int. Conf. on Electrical and Computer Engineering (ICECE), pp. 858-862, 20 Dec 12.

[3] E. Sharifahmadian and S. Latifi, "Advanced Hyperspectral Remote Sensing for Target Detection," IEEE 6041562, 21st Int. Conf. on Systems Engineering, Las Vegas, NV, 16 Aug 11.

[4] I. Leifer et. al., "State of the Art Satellite and Airborne Marine Oil Spill Remote Sensing: Application to the BP Deepwater Horizon Oil Spill," Remote Sens. Environ. 124, 185 (2012).

[5] T. Slonecker et. al., "Visible and Infrared Remote Imaging of Hazardous Waste: A Review," Remote Sens. 2, 2474 (2010).

[6] S. Trehan, "Reflection Seismology," Physics 240, Stanford University, Fall 2013.