When converting hyperspectral data into reflectance, a white reference is needed. It measures the incoming light seen by the hyperspectral camera, taking into account the optics’ transmission and the detector’s Quantum Efficiency. The role of white reference is crucial in obtaining good reflectance data with a hyperspectral instrument. This article highlights the importance of adjusting the height of the white reference tile when making measurements.
Placement of the white reference tile
Typically the white reference tile is placed at the same level as the top surface of the sample. However, if the sample is not flat, the top surface of the sample is not the most relevant position for the white reference. Instead, in this case, it is recommended to place the white reference to the middle “height” of the visible sample.
How does the placement of the white reference tile affect the sample measurements?
We conducted the following experiment to assess how the white reference tile level may affect spectra measurements.
- Specim FX17 hyperspectral camera was mounted on a lab scanner with dual illumination (see Figure 1).
- The top surface of the white reference tile was placed 20 mm above the black sample tray. The illumination of the scanner was adjusted so that it maximized at this white reference level.
- Two flat samples of plastics, PA and HIPS, were measured 13 times, every time at a different level to the white reference: -20 mm, -15 mm, -10 mm, -5 mm, the same level, +5 mm, +10 mm, +15 mm, +20 mm, +25 mm, +30mm, +35mm and +40 mm.
- The illumination was not adjusted, but it was kept constant for each measurement to maximize the white reference signal. We chose this approach because it is the most realistic scenario for using the system.
The analysis was carried out in two parts. Firstly, the samples were located lower than the white reference tile. Secondly, the samples were located higher than the white reference tile.
Figure 1: Specim lab scanner and the samples used in this measurement
1. Samples placed lower than the white reference tile:
Spectra of the samples were scrutinized and compared in Figure 2:
- The relative position of the samples with the white reference tile does not significantly affect the shape of the spectra. The change is mostly on their level.
- Regarding the calculated reflectance level changes concerning sample and white reference tile positions, it decreases while the sample gets deeper. The drop is significant as some wavelengths can be up to a few %.
Figure 2. Spectra of PA and HIPS acquired with Specim FX17 at different negative levels regarding the white reference position.
2. Samples placed at a higher than the white reference tile:
The results are illustrated in Figure 3:
- As observed in the previous section, the shape of the spectra is not affected, but only its level is.
- Here again, the calculated reflectance values vary with the level of the sample. This time there is no clear trend as observed previously, but the values increased or decreased with the increasing height difference. The variation can be more significant than 10% on an absolute scale, which is very substantial.
Figure 3. Spectra of HIPS acquired with Specim FX17 at different positive levels concerning the white reference position.
For both cases, the shape of the spectra is not affected by the difference in samples’ height with the white reference tile, but the spectrum level is. This is due to the variation of illumination irradiance along the pass-line of the camera.
Figure 4. illustrates when the light is well-adjusted to a single area and when the light is not well-adjusted to a single area. In the first case, the light irradiance will decrease when moving away from the 0 position, whereas, in the second geometry, variations are not only in one direction.
Figure 4. Light intensity variation along the pass-line of the camera depending on the illumination adjustment.
White reference title and sample level
For some applications, the absolute signal level of the spectrum is crucial. The signal level needs to be carefully considered for quantitative applications, as a bias in observed reflectance would yield significant misinterpretation of data. There are pre-processing tools available that can correct drift or offsets in spectra.
However, some applications require a precise adjustment of the white reference tile with the sample height, like colour measurement. In such cases, the samples and the white reference tile should be placed at the same level as precisely as possible.
Placement of the white reference is not as critical in applications that rely more on the shape of the spectra rather than the level. Sorting is one example. There could be a difference in heights, which allows for sorting samples with a particular profile, like wastes.
In the above example, HIPS and PA could still be sorted, even though they would not be placed at the same level as one of the white reference tiles. Samples could also have a different height, which would not affect their separation.
The Depth of Field should also be addressed. Samples that are not flat may be imaged blurry. This is not directly linked to the white reference tile height difference. However, the focus is typically adjusted at the same level as the white reference.
From an optical perspective, since the distances between the camera – samples and camera – white reference tile would not remain constant, it may be intuitive and straightforward to consider it as a reason for the spectral intensity variations. However, this is not correct.
The energy of a light ray indeed decreases with the square of its length. However, it’s compensated by the increase of the sample’s pixel size, which, consequently, also increased by an identical square factor, the area of which the light is collected.
Nevertheless, when the sample height varies, the geometry camera –sample – illumination differs as well, and this, combined with the reflective properties of the sample surface, may also explain the results of this study. If a sample has a perfect diffuse surface (Lambertian), then the reflected radiance measured by the camera is not affected by the change in the measurement geometry. But if there is any specular component in the sample’s reflectance, this may contribute to the measured variations. In this study, the samples are not Lambertian.
This technical note is prepared by Specim, Spectral Imaging Ltd. and is for general guidance only. We keep all the rights to modify the content.