we are interested in assessing plant growth/quailty parameters with a multispectral camera on an UAV and need calibration.
In cases that the 99% reflectance spectralon panel is saturated in the image.
It is suggested to use the panel that reflects more radiation but not saturated in the image.
We often use calibration targets in the form of tarps that are laid in the field during overflight. I recommend a series of 3-5 targets within the spectral range of the features in the image (i.e., from roughly 5%-50% reflectance for a plant-soil scene). This permits development of linear equations between the known tarp reflectance values and the digital numbers (DNs) from the tarp pixels in the drone image. Separate linear equations can be developed for each band of a multispectral or hyperspectral camera. For image calibration, the linear equations are then used to compute reflectance from image DNs over the entire image. Drone images have high enough spatial resolution to envision using a series of grey Spectralon plates on a level table, instead of field calibration tarps, though we haven't tried this yet.
To take it one step further, we also often measure the reflectance values of the tarps during overflight using a field spectroradiometer. A 99% Spectralon panel is used as the reference for computing spectral reflectance of each tarp from spectrometer data. A similar approach could be used to measure the spectral reflectance of other static features in the scene during overflight (roads, bare soil, etc.) for comparison to that in the calibrated drone image.
It is easier to make recommendations if you state whether you want to place a calibration target in-scene (imaged during flight) or before/after a flight.
For in-flight, if your camera has automatic gain control (AGC) it will very likely saturate over a bright target unless it is huge (like the kind Kelly has - I'm guessing Tracor Aerospace chem-treated canvas, gray ones are usually safe).
You only need to calibrate the range of reflectance values you want to analyze. Therefore, only risk using a white spectralon if you are expecting NIR reflectance upwards of 80%. Visible bands shouldn't be nearly that much but I'm not sure if something like flowering radish could creep up there.
Just as important as spectralon color: if you do use AGC be sure to adjust for gain, exposure, and vignette BEFORE reflectance compensation and make sure that is all done before mosaicing (if you are satisfied with "orthomosaics" derived by structure from motion (SFM)).
NIST-traceability is sometimes overkill precision after you factor in other errors such as variation in light source intensity so I wouldn't shell out extra money for that (the calibration curve within the range of a silicon sensor is typically 0.99). If you do use a SWIR sensor then disregard this paragraph.
We use 1.2 x 1.2 m fabric reference panels (VIS-NIR range) shot at beginning and end of a flight from the air.
(The average reflectance of a panel in the scene changes as the UAV increases altitude, and seems to stabilise above about 30 m for several multispectral and adapted consumer camera models that we use.)
We have found that targets with greater than 60% reflectance may saturate. (Approximate values. These change depending on the camera sensitivity/dynamic range and settings, and therefore the reflectances obtained in the different bands by the particular camera.)
We have good results using 2 or 3 targets in the range 10% to 50%. If using only one target, I prefer a target within the 40-55% range.
when moving closer to the fruits within the canopy, we use even 80% (uncalibrated) spectralon
- Badges
- Science topic
- Similar topics
- Biological Science
- Zoology
- Ornithology