The versatility and precision of hyperspectral imaging make it an indispensable tool in numerous scientific and industrial applications, from medical imaging to environmental monitoring to quality control. But traditional hyperspectral imaging systems can be costly, cumbersome, and challenging to scale. A computational spectral imaging system from the University of Utah provides a fast, inexpensive, efficient alternative to capturing high-quality spectral data. The system, which the team tested across biomedical, food-quality, and astronomical use cases, could establish a new framework for high-speed, high-fidelity spectral imaging with broad translational potential. The system uses a diffractive filter array to project spectral information into the spatial domain, enabling the capture of a single-channel, 2D image that contains both spatial and spectral data. This 2D image, called a diffractogram, is computationally decoded to reconstruct a spectral image cube with 25 spectral bands in the 440-800 nm range. Each of the 25 separate images that comprise the cube represents a distinct slice of the visible spectrum. The team modeled and designed the diffractive filter array and the algorithm to reconstruct the hyperspectral images from the raw data captured by the sensor. By encoding a scene into a single, compact 2D image rather than a massive 3D data cube, the camera makes hyperspectral imaging faster and more efficient. The fast encoding enables the system, which is small enough to fit into a cellphone, to take high-speed, high-definition video. Researchers at the University of Utah have developed a computational spectral imaging system with a diffractive element that divides light into 25 different color channels, instead of three. Courtesy of the University of Utah. “One of the primary advantages of our camera is its ability to capture the spatial-spectral information in a highly compressed, two-dimensional image instead of a three-dimensional data cube, and use sophisticated computer algorithms to extract the full data cube at a later point,” professor Apratim Majumder said. “This allows for fast, highly compressed data capture.” The current prototype camera can take images at just over one megapixel in size (1304 x 744 pixels) and break them down into 25 separate wavelengths across the spectrum. The diffractive element, placed directly over the camera’s sensor, encodes spatial and spectral information for each pixel on the sensor. “We introduce a compact camera that captures both color and fine spectral details in a single snapshot, producing a ‘spectral fingerprint’ for every pixel,” professor Rajesh Menon said. To demonstrate the camera’s capabilities, the researchers applied standard inferencing techniques to reconstructed spectral images across various sectors. The system demonstrated a spectral reproduction error of less than 15% across the 440-800 nm band. The researchers used the camera to classify lung and trachea tissues in ex vivo chicken lung images, predict the freshness of strawberries, and mimic the spectral filters that are used in stellar imaging. These experiments highlighted the system’s potential in medical diagnostics, food-quality assessment, and astronomical observations. The diffractive, computational spectral imaging system offers several advantages. It provides snapshot capability, eliminating issues with scan-and-stitch methods. The diffractogram serves as a form of optical compression, efficiently storing spatio-spectral content in a compact, information-rich 2D array. This is particularly beneficial for applications with limited storage or transmission bandwidth, such as airborne or satellite imaging. “Satellites would have trouble beaming down full image cubes, but since we extract the cubes in post-processing, the original files are much smaller,” Majumder said. The system also provides the flexibility to perform reconstructions offline and on-demand, after data capture, for scenarios with limited on-board computational resources. Also, since the diffractogram encodes spectral information continuously, it allows for information to be selected on an application-specific spectral basis, yielding smaller image cubes and faster, more stable reconstructions. Compared to traditional hyperspectral imaging systems, the computational spectral imager’s streamlined approach reduces costs significantly. “Our camera costs many times less, is very compact and captures data much faster than most available commercial hyperspectral cameras,” Majumder said. “We have also shown the ability to post-process the data as per the need of the application and implement different classifiers suited to different fields such as agriculture, astronomy, and bioimaging.” The diffractive element used in the new spectral camera contains a repeating pattern of nanoscale features. Courtesy of the University of Utah. Hyperspectral cameras have long been used in agriculture, astronomy, and medicine, where subtle differences in color can make a big difference. But these cameras have historically been bulky, expensive, and limited to producing still images. “When we started out on this research, our intention was to demonstrate a compact, fast, megapixel-resolution hyperspectral camera, able to record highly compressed spatial-spectral information from scenes at video-rates, which did not exist,” Majumder said. “This work demonstrates a first snapshot, megapixel, hyperspectral camera,” he said. “Next, we are developing a more improved version of the camera that will allow us to capture images at a larger image size and an increased number of wavelength channels, while also making the nanostructured diffractive element much simpler in design.” By making hyperspectral imaging cheaper, faster, and more compact, the computational camera advances spectral imaging technology and potentially opens the way for technologies that could change the way the world and its contents are seen. The research was published in Optica (www.doi.org/10.1364/OPTICA.559279).