TECHNOLOGY
What is Hyperspectral Imaging?
Color imaging was originally developed to replicate human visual color perception, which consists of three different color sensitivity functions in our retina. This results in three distinct color channels: red, green, and blue, as illustrated on the left in the figure below. Compared to conventional RGB color images, spectral images can provide higher-dimensional color information, such as spectral reflectance, which reveals the intrinsic chemical properties of an object's surfaces.
Spectral images can be categorized into two main groups: multispectral and hyperspectral images. This classification reflects a historical grouping. Multispectral images are captured using a multispectral camera, producing five to ten spectral channels for the visible spectrum with a rotating bandpass filter wheel. In contrast, hyperspectral images are obtained through a scanning-based spectral camera, generating 30 to 60 channels for the same spectral range, which is five to ten times more than that provided by multispectral images. Hyperspectral images provide much denser information than ever before, not limited to the human visible spectrum but extended to the invisible spectrum, such as the short-wavelength infrared spectrum. Our product line embodies this hyperspectral imaging technology, providing dense spectral information from visible to invisible infrared wavelengths.
RGB image
Multispectral image
Hyperspectral image
What Spectral Range is Captured?
The spectral wavelength range humans can perceive spans from 0.4 to 0.7 μm and is represented as three channels: red, green, and blue. Each channel covers approximately 0.1 μm, a range that is too broad for scientific and engineering inspections using machine vision in automation machinery manufacturing. As shown in the figure above, hyperspectral imaging can capture denser spectral channels than other imaging methods, expanding the spectral range beyond the capability of the human visual system. In particular, the infrared (IR) spectrum, including near IR and short-wavelength IR, is very effective for food and recycling applications in computer vision. Our cameras continuously cover the range from 0.4 to 1.0 μm with high accuracy and speed. Hypergram Inc. plans to gradually extend the spectral range of its cameras to include the SWIR region up to 1.7μm.
What Hyperspectral Imaging Technologies are Available?
There are three popular types of cameras used in hyperspectral imaging technology: filter-based, push-broom, and compressive cameras. A filter-based hyperspectral camera consists of either a bandpass filter wheel or a liquid crystal tunable filter (LCTF). The spectral filter changes its transmittance sequentially to scan the entire target spectrum range. The push-broom camera operates in a similar manner. While the filter-based camera scans along the spectral axis of the captured spectral volume, the push-broom camera line scans in a single direction of an image. Both filter-based and push-broom cameras provide high accuracy; however, the objects or scenes must remain static during scanning. Consequently, the applications of these technologies have been limited to scanning static objects, conveyor belt imaging, or aerial imaging. It is not surprising that capturing hyperspectral video with industry-level accuracy is far beyond the current capability of state-of-the-art hyperspectral imaging. Compressive spectral imaging was proposed to enable snapshot hyperspectral imaging. However, it suffers from low image quality and low spectral accuracy and has thus been extensively experimented only in scientific laboratories.
Filter-based camera
Push-broom camera
Compressive camera
Advanced Technology by Hypergram Inc.
Our team has developed a new hybrid hyperspectral imaging technology by taking advantage of the existing hyperspectral imaging technologies to address the longstanding trade-off between accuracy and performance in capturing hyperspectral image data. Our research team at KAIST Visual Computing Lab has extensively explored compressive hyperspectral 3D imaging, prism-based snapshot spectral imaging, and diffraction-based snapshot hyperspectral imaging, which we have presented at prestigious computer science conferences such as ACM SIGGRAPH over a decade. By building a KAIST startup company, Hypergram Inc., we have challenged ourselves to invent a new, practical yet advanced technology for industry applications by reformulating our scientific elements to bring the most accurate and fastest hyperspectral imaging technology practice ever to market as a commercial product through the innovation of AI technology. Here is an example of hyperspectral images captured by one of our hyperspectral imaging technologies.
397 nm
464 nm
480 nm
497 nm
516 nm
560 nm
Hyperspectral image
577 nm
604 nm
635 nm
732 nm
791 nm
916 nm