Hyperspectral Sensors, a Quick Guide

The hyperspectral cube

Hyperspectral sensors capture multiple narrow band images over a spectral range. While common RGB images have just three chromatic components, hyperspectral images feature many spectral bands, with a spectral dimension added to the two spatial dimensions characteristic of images. As a result, they’re usually referred to as hyperspectral cubes.

As they contain information at different specific wavelengths, hyperspectral cubes enable the analysis and identification of objects based on their spectroscopic properties.

Types of hyperspectral sensors

There are three main methods of acquiring a hyperspectral cube: spatial scanning, spectral scanning, and snapshot imaging. They’re based on different sensor arrangements and acquisition principles. Each option results in a different tradeoff between spatial resolution, spectral resolution, and acquisition time.

Snapshot hyperspectral sensors

The last trend in the field, snapshot sensors collect the entire 3D datacube in a single shot without scanning. As a result, they don’t need any mechanical part and are free of scanning artifacts. They’re far more robust, compact, and cost effective than any other HSI sensing approach and have higher optical efficiency.

The best example of snapshot sensors are IMEC’s patented mosaic sensors that feature a mosaic of filters deposited or placed on top of standard sensors. Mosaics can be designed on purpose in different arrangements like 4x4 pixels (16 filters) or 5x5 pixels (25 filters). While they tradeoff spatial and spectral resolution, they’re the only option for real time snapshot hyperspectral imaging, a killing advantage for so many applications.

We may find them in a great camera brand like Photonfocus for visible and NIR range. SpectralEdge provides a seamless connection and acquisition process with these amazing HSI sensors, allowing to take the most of their impressive 300 fps capability in e.g. a high speed product inspection and sorting use case.

Spatial Scanning sensors

Spatial scanning is one of the traditional methods used for acquiring hyperspectral cubes. In this approach, many spectral bands are acquired for a single spatial line at a time by using a dispersive element like a prism or grating. The remaining spatial dimension is scanned through different approaches, such as using a moving mirror, a rotating prism or through camera or object movement. As a result, they are more prone to motion artifacts if there are movements or vibrations during the scanning process.

Spatial scanning cameras have been widely used for years and are known for their high spectral and spatial resolution. The scanning process can be time-consuming, especially for high-resolution images, which usually prevents real-time applications. The use of a dispersive element and even mechanical components make them bulky and expensive. However, they’re still the sensor of choice when spectral and spatial resolution are a must.

Spectral scanning sensors

Rather than scanning the spatial dimensions, spectral scanning sensors acquire the entire scene simultaneously. They usually achieve this by using a tunable filter to selectively transmit different spectral bands onto a 2D detector array in relatively short times. While they are often much faster than a spatial scanning sensor they’re not truly snapshot and may still suffer from motion artifacts.

In the end, spectral scanning sensors trade-off their spatial and spectral resolution, as the performance of the tunable filters can affect these parameters.

Acquiring and handling the hyperspectral cube

When starting to work with hyperspectral sensors one major hurdle is the extraction of useful information from the sensor. First of all, we need to acquire the images. Next, we need to reconstruct the hyperspectral cube and finally we need to analyse it. Each of these steps require specialised software routines, that are usually hard to craft. Moreover, HSI images may be too heavy to handle in a consumer PC.