Technical

Understanding Raw Data, Radiance Data, and Reflectance Data in Hyperspectral Imaging

By Dr. Alexandre Lussier, Application Engineer - October 28, 2024

Hyperspectral Imaging Data: Overview

The following summarizes the differences between — and uses of — the various types of hyperspectral data.

Hyperspectral Raw data:

Dimensionless, uncalibrated (except spectrally), and instrument-specific
Represented as digital numbers (DNs) or counts
Can not be compared with data taken by other instruments, under different illumination conditions, or using different acquisition parameters, until those data are further processed to account for those differences.

Hyperspectral Radiance data:

Measured in the physically meaningful units of microFlicks (power per unit solid angle per unit area)
Related to the energy emitted or reflected from the object being scanned
Data depends on atmospheric effects (scattering, absorption) and changes in illumination conditions (e.g., sun angle)

Hyperspectral Reflectance data:

Represents a physical property of the object, independent of illumination conditions or hyperspectral instrument differences
Allows for direct comparison between data taken in different measurement conditions or with different hyperspectral instruments
Ideal for material identification and analysis

The Details Behind Raw Data, Radiance Data, and Reflectance Data in Hyperspectral Imaging

Before discussing hyperspectral data types, we will review the components that come together to create hyperspectral data. Each of these components has an impact on the data captured by the hyperspectral imager.

Hyperspectral imaging requires a:

Broad-band source of electro-magnetic radiation (i.e. illumination)
Object(s) to scan
Hyperspectral imager

An example is shown in the following figure where a hyperspectral imager is monitoring a tree illuminated by sunlight.

Figure 1: Example showing illumination source, object being scanned, hyperspectral imaging sensor and hyperspectral imaging data.

Illumination

All forms of broad-band illumination, whether solar or human-made, have varying intensity as a function of wavelength. See the plots of solar illumination and two of our benchtop broadband light options below. In Figure 1, the illumination strength as a function of wavelength is denoted I(λ).

Figure 2: Radiance of Spectralon® illuminated by solar radiation.

Figure 3: Radiance of Spectralon® illuminated by two different Resonon benchtop lighting systems.

Object Being Scanned

The chemical composition of an object governs its interaction with various wavelengths of incident light, determining which wavelengths of light are absorbed, transmitted, or reflected; measuring this interaction is usually the purpose of hyperspectral imaging. In Figure 1, the reflected signal collected by the hyperspectral imager depends on how much light is incident on the tree, I(λ), and the reflectance, R(λ), of the tree. The wavelength-dependent reflectance changes from object to object. This is why different objects appear as different colors to our eye, and why different objects can be identified using hyperspectral imaging.

Hyperspectral Imager

The hyperspectral imager itself has varying efficiency at different wavelengths. An example of that is the Pika IR+ spectral response plot shown below. In Figure 1, the spectral-dependent response function, which depends on the detector efficiency, is denoted as H(λ).

Figure 4: Spectral response plot of Pika IR+ showing wavelength dependence.

Each of the terms shown in Figure 1 affect the recorded hyperspectral data. Next, we will explain how each of the different hyperspectral data modes includes or removes these effects.

Hyperspectral Raw Data: The Beginning

Every hyperspectral datacube begins with Raw data. Raw data is the product of all the terms mentioned above, represented in Figure 1 as I(λ)⋅R(λ)⋅H(λ). These Raw data depend on the wavelength of the input light and are in units of digital numbers, or DNs (sometimes referred to as "counts"). DNs are proportional to the energy incident on the pixel but are dimensionless and uncalibrated units.

The DN for each pixel ranges from 0 (less energy on that pixel than the minimum energy the sensor can resolve) to the maximum value (fully saturated), which is a function of the bit depth of the sensor. For instance, a 12-bit sensor like the Pika L would have a maximum DN of 4095 (2^12 possible digital numbers, starting from 0).

Figure 5: Representation of a hyperspectral datacube.

Resonon’s hyperspectral imaging software packages automatically subtract the hyperspectral camera’s average dark current from the Raw data, provided a Dark Correction Frame was collected. More information about Dark Current Noise is available in our post on SNR. A simplified formula for Raw data is: \[Raw\,Data = {DNs}\]\[Dark\, Corrected\,Data = {Raw\,Data - Dark\,Correction}\]

Raw data represents the original sensor output, enabling later reprocessing or conversion to other data types. Raw data cannot be directly compared between different instruments, nor even between datasets from the same instrument if there are any changes in illumination or measurement configuration.

Hyperspectral Radiance Data: Quantifying Energy

Radiance is a unit commonly used for scientific studies with units of power per area, bandwidth, and solid angle, and is related to the amount of electromagnetic energy collected by the hyperspectral imaging sensor. In Figure 1 Radiance is represented by the term I(λ)⋅R(λ).

Radiance data is derived from Raw data through hyperspectral imager calibration, converting the dimensionless units of DN into physical units. Resonon, and many other hyperspectral imaging companies, provide Radiance data in units of microFlicks: \[1\,\mu Flick = {{1\, \mu Watt}\over{steradian \times cm^2 \times \mu Meter}} \]

Radiance data is valuable for studies that seek to quantify the energy collected by the sensor, such as in atmospheric research or when comparing data from different sensors. However, for many applications, the atmospheric influence and illumination variability that are contained with Radiance data can be a hindrance.

Figure 6: Airborne radiance hyperspectral image with the spectra of several different material types.

To generate Radiance data, a radiometric calibration is performed on the hyperspectral imager. This creates an Imager Calibration Pack (ICP file), which can be used along with appropriate software plugins, to convert Raw data into Radiance data. The ICP file is unique to the particular hyperspectral camera and objective lens from which it was created. \[Radiance = {(Raw\,Data - Dark\,Correction^{\ast}) \times Pixel\text{-}by\text{-}Pixel\,Gain^{\ast}} \] \[^{\ast}\,{Both\,of\,these\,terms\,are\,from\,the\,ICP\,file} \]

Hyperspectral Reflectance Data: Surface Properties Revealed

Reflectance data frequently proves to be the most useful type of spectral data as it solely depends on the object of interest. In Figure 1, Reflectance is represented by the term R(λ). Reflectance is the ratio of the amount of light reflected from an object to the amount of light incident on the object. Transmittance, which is similar to Reflectance, is the ratio of light transmitted through an object to the amount of light incident on the object. As ratios, both are unit-less quantities, generally displayed as a percentage or as an integer value (Spectronon uses 0-10,000).

Reflectance and Transmittance data represent physical properties of the object being imaged. These data are independent of the illumination, atmosphere, and hyperspectral imager used. Because of this, these data types are frequently used when attempting to identify an object via spectral imaging.

Figure 7: Reflectance data curves of some different plant species (image courtesy of www.microimages.com).

To derive Reflectance data, it is necessary to remove the hyperspectral imager response and the illumination spectral distribution from the Raw data. This is commonly accomplished in one of the following ways:

For benchtop systems, scanning a reflectance standard (a calibration tile) made of materials like Spectralon®, Fluorilon®, or Teflon. This is sometimes called a “white reference”. The materials above have a nominal reflectance of 1.0.
For outdoor or airborne systems, using a ground calibration tarp with known spectral reflectance placed in the scene being scanned. The known reflectance of the tarp is used to deduce the incident light distribution and calibrate the rest of the scene.
For airborne data collection, it is possible to use a downwelling sensor mounted atop the airborne vehicle to collect solar spectral irradiance continuously during flight. Using spectral irradiance and calibration data of the instrument itself enables one to determine object reflectivity.

As an example, the simplified formula for Reflectance with a hyperspectral benchtop system is as follows:

\[Reflectance = {{Raw\,Data - Dark\,Correction} \over {White\,Reference - Dark\,Correction}}\]

Conclusion: Summarizing Hyperspectral Data Types

Understanding the different types of hyperspectral imaging data is essential for effective data collection, analysis, and data interpretation.

Raw data preserve the original sensor information and allow for reprocessing
Radiance data are related to the energy reflected or emitted by an object
Reflectance data reveal properties of the chemical composition of the object being scanned and thus, facilitate material identification

Each data type serves a specific purpose in the hyperspectral imaging workflow, from initial data capture to final analysis. By understanding these distinctions, researchers and practitioners can choose the most appropriate data type for their applications, whether in environmental monitoring, geological surveys, agricultural assessments, or material analysis via spectral sensing techniques.

If you have questions or if you would like to better understand how hyperspectral imaging could help you solve challenges in your industry or research, please reach out to our Sales Team.

Dr. Alexandre Lussier, Application Engineer

Alexandre Lussier is an application engineer at Resonon. His work involves helping professionals from various disciplines understand hyperspectral imaging and how it can resolve scientific questions and industrial challenges.

Alex holds a PhD in physics with expertise in material science, spectroscopy, and education. He has spent half his professional career in academia and half in industry, including work in fuel cells, construction, solar power systems, and fine carpentry.

His personal pursuits include summer and winter human-powered mountain adventures, surfing, and art. Whatever your potential application, he is curious to hear about it.