Question: Mapir Survey3 RGN image processing pipeline

mrodriguezorejuela asked on September 28, 2018 13:45
141 views | 1 answers | #17185


Hi there. I need some input regarding how to process some images from our Mapir Survey3 camera in some very specific conditions.

General info: the camera collects red, green and nir channels and is set in an incubation chamber pointing at the plants. The light in the chamber is very stable, intensity wise, but it's 'redder' than the usual fluorescent.

Now, if I wanted to calculate the NDVI for a given picture, i would need to calculate (NIR - VIS) / (NIR + VIS) but in this case i don't have visible. I've seen that some people use the red channel as visible since there's the notion some nir also gets into the green channel.

Given these parameters, would you consider calculating the NDVI from the red and nir channel okay? I am concerned regarding having a redder light input, as the red channel pixel intensity seems very high compared to the blue one (which is very sparse). Should I be concerned about not capturing enough NIR?

Thanks in advance, hope to get some discussion started here!



3 Comments

the camera collects red, green and nir channels

So this camera has a red filter and captures NIR in the blue channel.

The light in the chamber is very stable, intensity wise, but it's 'redder' than the usual fluorescent.

The standard interpretation of NDVI values assumes that the photos were taken with the plant foliage illuminated by sunlight. If your plants are illuminated with artificial lights or sunlight filtered through modern windows, the ratio of NIR to visible light could differ enough to make NDVI meaningless.

would you consider calculating the NDVI from the red and nir channel okay?

Yes, NDVI = (NIR-R)/(NIR+R) is the standard way to compute NDVI. Both the red and green channels will be contaminated with NIR. If there is data on which channel is less contaminated, that might support using a different formula.

the red channel pixel intensity seems very high compared to the blue one (which is very sparse). Should I be concerned about not capturing enough NIR?

Yes, you should be concerned. Consumer digital cameras are not very sensitive to NIR and the ratio of Red:NIR in the photo will generally not be anywhere close to the Red:NIR ratio in the photographed scene.

Mapir supplies a calibration target, calibration software, and instructions for how to use them. This calibration procedure or some other procedure must be followed to produce NDVI images that might have biological meaning. The primary adjustment made by the calibration procedure is to exaggerate the intensity of the NIR channel. Some calibration procedures might also reduce the intensity of the visible channel to compensate for the contamination of that channel with NIR.

Chris

Is this a question? Click here to post it to the Questions page.

@cfastie Thanks for your detailed explanation, that helped a ton! I do have the calibration targets from Mapir, unfortunately I need to assemble our own calibration system as theirs is not open and not useful for automated processing :)

If you don't mind, let me ask you one more question: is there a systematic way to evaluate channel bleeding? I would like to see if and how much of it happens with my cameras.

Is this a question? Click here to post it to the Questions page.


That's too bad about the proprietary Mapir calibration. Is there information about the spectral reflectivity of the calibration targets? If not, those targets will not be of much use to you unless you can measure their spectral response.

The Photo Monitoring plugin for FIji has a calibration routine. It requires that you have targets of known spectral reflectivity to put into the photos.

Learning about the amount of each wavelength that is captured in each channel of a camera is not easy. There are data about this for certain cameras, but there is very little data for cameras after they have been modified by removing the IR cut filter. There is a discussion of the Canon PowerShot A2200 which was the subject of a study in which they measured the spectral response of the camera before modification, with the IR cut filter removed, and then with a red cutoff filter (like the one in your Mapir).

To create similar data on spectral response of a camera, some fancy equipment is required. You could get a cruder version of this information by illuminating a scene with individual LEDs of known emittance. There are inexpensive LEDs with narrow spectral ranges (e.g., all red, all blue, or all NIR). So if you can trust the LED to emit only NIR, then you can illuminate a white surface, take a photo, and see if any light is captured in each camera channel (R,G,B). It is very difficult to quantify the intensity of light in each channel unless you can capture RAW image files and you know how to use them (jpeg photos produced by a camera will have been subjected to various manipulations including white balance and gamma correction).

If you don't trust the LED specifications, you can check them with a simple spectrometer like the ones sold by Public Lab. That would require that you learn how to check the specifications of the spectrometer.

Careful inspection of the detailed spectral response graphs of the PowerShot A2200 in the article above can allow a good estimate of how to modify the photo data to compensate for NIR or UV contaminating a channel used for visible light data, or for visible light contaminating the NIR channel. Using a DIY technique with LEDs may or may not allow such meaningful estimates.

Chris

The links cited above are listed below because the edit window was not allowing link insertion or preview.

https://publiclab.org/notes/nedhorning/01-13-2016/packaged-photo-monitoring-plugins-available-on-the-github-repositoy

https://groups.google.com/forum/#!searchin/plots-infrared/article|sort:date/plots-infrared/aJhM30D6bUM/zuZqlFYiHQAJ

https://www.researchgate.net/publication/281464426_Estimation_of_the_spectral_sensitivity_functions_of_un-modified_and_modified_commercial_off-the-_shelf_digital_cameras_to_enable_their_use_as_a_multispectral_imaging_system_for_UAVs

Is this a question? Click here to post it to the Questions page.


Reply to this comment...


Log in to comment