The Role of Spectral Resolution in Hyperspectral Imaging Applications

But whether it is the photoreceptor cells in our retina or the Bayer filter on a camera sensor, grouping visible light into three broad channels discards much of the information it holds.  

Light interacts with materials in many different ways. This leads to spectral signatures in reflected light that can identify and reveal specific properties of the materials present as well as various chemical processes. It is possible to capture these spectral signatures by imaging the world in more detail, going beyond the limitations of the human eye and RGB cameras. 

This is the goal of Hyperspectral Imaging (HSI), using a larger number of narrower channels to recreate the spectra of light throughout the image and reveal new information. Already used across a number of fields, HSI has the potential to transform industries around the world: 

• Agriculture: Real-time monitoring of plant science metrics to optimise crop yields and resource use. 
• Environmental monitoring: Measuring and tracking the impact of human activity, large-scale vegetation monitoring, mineral exploration, and more. 
• Healthcare: Revealing biological properties for use cases in diagnostics, surgery, and monitoring. 
• National security: Enabling military and intelligence personnel to understand their environment better and uncover hidden threats. 

The suitability of hyperspectral imaging for a specific application depends on many factors. Some of these are more practical in nature (robustness, portability, etc.), and some are based on performance. The hyperspectral camera needs to image in enough detail to detect the spectral signatures of interest for a given application. Its ability to do this is defined by the system’s spectral resolution. 

What is spectral resolution?  

Spectral resolution is a key parameter for any instrument that measures the intensity of light across different wavelengths. It defines a spectrograph’s ability to distinguish between two wavelengths. High spectral resolution refers to devices with a larger number of narrow bands, while low spectral resolution means fewer, broader bands. 

The spectral resolution of a hyperspectral imager is defined by: 

• Spectral range: The range of wavelengths that the sensor can observe. Normally expressed in nanometres (nm), the spectral range defines the types of signatures that can be analysed. Many hyperspectral cameras cover the visible and near-infrared (NIR) range to identify a wider range of materials and phenomena. 
  
• Number of spectral bands: The number of channels the camera uses to group and record light. A larger number of spectral bands typically means higher spectral resolution. For example, black and white photography only has one band, and traditional (RGB) cameras only have three. In contrast, hyperspectral systems have tens or hundreds of channels to enable the identification of spectral signatures. 
  
• Width of the spectral bands: The size of each individual band, generally measured in nm. Higher spectral resolution means smaller band widths, i.e., the camera more precisely records the wavelength of light hitting the sensor. The width of bands often varies across the spectral range of the camera. Therefore, we cannot simply divide the spectral range by the number of bands. 

Hyperspectral resolution is often quoted in nm alongside the associated spectral range. For example, a VNIR (visible/near-infrared) hyperspectral camera might quote a spectral resolution of 5nm between 450-700nm and 10nm between 700-900nm. 

While an RGB camera delivers three values for each pixel to output colour images, hyperspectral systems divide light into more channels, returning more values across the image. This is often referred to as a hyperspectral data cube, with an x and y axis representing the spatial information (pixels) and an additional λ axis representing spectral data (intensity of light across each spectral band). 

People are more likely to associate the word “resolution” with pixel size. But while the size of the pixels along the x and y axis defines the spatial resolution of the hyperspectral camera, the number and width of channels in the third dimension (λ axis) define its hyperspectral resolution.  

A larger number of narrow bands allows the camera to more accurately recreate the incoming spectra of light and, therefore, learn more about the materials and chemical processes within the scene. 

Spectral resolution and hyperspectral imaging performance

Spectral resolution defines the performance of a hyperspectral imaging system in terms of identifying signatures and learning new valuable information about the image. Higher spectral resolution means the camera outputs continuous spectral curves across the image, revealing vital information about the objects and phenomena present. 

This allows hyperspectral imaging systems to perform a range of functions: 

• Identifying materials: Detecting spectral signatures related to different materials. 

• Distinguishing between similar materials: Capturing reflectance spectra with enough detail to identify subtle differences between similar materials. 

• Spotting changes over time: Detecting small spectral changes over time to identify the underlying processes at work. 

• Improved accuracy and precision: Making mistakes in spectral analysis less likely. 

In an ideal world, hyperspectral cameras measure the exact wavelength of light, recreating the true spectra of light across an image. However, in the real world, there are technological, practical, and performance trade-offs to consider when incorporating hyperspectral imaging. 

Firstly, higher spectral resolution increases data requirements, i.e., the camera has to output a considerable amount of data (imagine the λ-axis of the hyperspectral data cube getting larger and larger) for every frame. Handling large volumes of data leads to significant storage, process, and analysis challenges, requiring more compute and slowing down the maximum readout rate of the camera.  

This can limit edge use cases where the compute available is reduced and real-time monitoring scenarios that require constant hyperspectral frames. Higher spectral resolution systems typically rely on line-scanned hyperspectral technology that pans across the scene, combining many outputs to build a single data cube. This significantly increases the time it takes to produce an image and can introduce motion artefacts when imaging a non-static scene. 

Next, spreading the same light input across more and more channels reduces the photons recorded for each. This lowers the signal-to-noise ratio, reducing the overall quality of the images being output. 

Finally, there are a number of practical issues using high spectral resolution systems. The technology required for top-end spectral resolution cameras raises costs, making them less accessible. They are also more complicated devices, often introducing additional optical elements to scan across the scene. This makes them less robust for measurements outside of the lab and harder to use without significant training. 

Choosing the right hyperspectral camera for an application

The spectral resolution required varies depending on the application, and given these trade-offs, it is vital to consider other factors.  

For example, doing open-ended exploratory analysis requires high spectral resolution as you do not know what you are going to find. However, developing a system with a fixed goal and a set number of spectral characteristics to investigate puts a ceiling on the spectral resolution required.  

A hyperspectral camera for agriculture and ecology may only need to determine certain spectral indices such as NDVI, chlorophyll content, and key stress factors (diseases, drought, etc.). A plastic recycling system has to distinguish between common plastics used in packaging. It does not need a detailed spectral analysis to identify a vast range of materials. 

Seeking to maximise spectral resolution at the expense of other performance parameters is often counterproductive. Assuming you choose a camera with spectral resolution above the floor for the given application, it is beneficial to ensure other factors are also met.  

The Living Optics video rate, snapshot hyperspectral camera

The Living Optics Camera provides video rate hyperspectral data powered by snapshot technology. With 96 bands over a VNIR spectral range of 440-900nm, the camera is optimised for light throughput and high frame rates, it offers spectral resolution with typical values of: 

• 8nm at 450nm 

• 15nm at 600nm 

• 22nm at 860nm 

Delivering detailed spectral analysis in real-time at frame rates up to 30Hz. Plus, with a robust, handheld design, the Living Optics camera offers portability and simplicity of use, enabling live measurements in the field with minimal training. 

Want to learn more about our cutting-edge hyperspectral technology and how it could be incorporated into your operations? Get in touch with our sales team today.