Have you attended an online call with Public Lab? We'd love your feedback!

# Plant Health NDVI Consumer camera vs Professional multispectral camera

by Claytonb | 09 Jul 17:50

Claytonb was awarded the Basic Barnstar by warren for their work in this research note.

I wanted to share some of my results in comparing a modified consumer camera with a scientific grade multispectral camera when evaluating plant health with NDVI. This note is related to NDVI only as the modified consumer camera tested only has two bands to compare with(Red and NIR). I wanted to see if it was possible to get meaningful and comparable NDVI data from an affordable consumer grade camera vs a scientific multispectral camera and if the difference in price was really worth it.

NDVI is a vegetation index that uses a numerical indicator using visible and near-infrared bands to help analyze remote sensing measurements. It is often used in agriculture to measure general crop health and changes. It is calculated by dividing the difference in the near-infrared and red color bands by the sum of near-infrared and red bands for each pixel.

The consumer camera used was the Canon S100 with the internal IR filter removed and replaced with the MidOpt DB 660/850 filter(http://midopt.com/filters/db660850/) MidOpt Band info

The scientific camera used was the Micasense RedEdge 3(http://www.micasense.com/rededge/) RedEdge Band info

Canon S100 Red and NIR channels

RedEdge 3 Red and NIR channels

Both sets of camera images were calibrated against reflectance panels with known values. Only RAW images were used. The images were obtained on the same day at around solar noon. The flight with the Canon camera was flown within five minutes of landing with the RedEdge camera. The image sets were processed with Pix4D Pro with the only difference being a slight modification to the NDVI formula with the Canon to account for the NIR contamination in the red channel.

Side by side comparison

The results are very similar especially in terms of sensitivity. Can you tell which image set came from a near $7000 camera and which came from a camera that cost under$200? The Canon set is on the left of the above image and the RedEdge set is on the right. With the advances and availability of new filters to modify a consumer camera I think the gap is closing. Scientific cameras certainly have their abilities and value but the modified consumer cameras are proving their abilities and usefulness as well.

Hi Claytonb, Really interesting analysis. I currently use a Canon S110 modified by Sensfly for NDVI and I wondered if there could be a big difference with a camera with separate sensors - I noticed now the accuracy that the camera can offer. Impressive! I also use Pix4D pro, what are the slight modification to the NDVI formula? (I imagine they will be different depending on the type of filter) And also, where did you find the calibration board? With known reflectance?

Thanks Alex

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

Hi Alex, I would recommend contacting Charlie Langlois at Labsphere- clanglois@labsphere.com They can make affordable(\$300 range)targets that are smaller 2X2" pieces that work great with same quality and accuracy as the targets that cost thousands. The formula I used in the above comparison subtracted around 40% of the pixel value in the red channel. I imagine it can be very different with different sensors and filters. It also changes based on the shutter speed, ISO as well so you will have to play with it to see what you can get. Luckily I had the RedEdge camera and a Greenseeker to verify values and know what I was aiming for in the formula modification.

I really liked this note too! I'm interested in how this dataset could be used as a "test set" for software development as well. There are a range of software packages (and methods using only Photoshop or Gimp) to do NDVI colormaps, and this is a great, well-documented and well-controlled set that could be used to validate the software procedures different packages use.

Thanks for posting such a great and clear note!

Thanks @Claytonb for this really interesting post. Could you detail a bit more about how you calibrated your images with reflectance targets please?

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

@everydayduffy -For the above comparison I used the calibration option in Pix4D pro. It consisted of drawing a region of interest on the image of the reflectance panel and inputting the known values for given wavelength. Ned Horning has an excellent plugin used with ImageJ that does a great job using calibration targets with known reflectance values. https://publiclab.org/notes/nedhorning/07-22-2015/introducing-the-calibration-plugin-for-imagej-fiji

Thank you @Claytonb.

@Claytonb, This is a great post and comparison. I have a couple questions, I'm using the V2 NoIR Raspberry Pi camera and #2007 filter they provide is a good place to start with this but really isn't ideal it seems if you want anything better than an estimation as it doesn't have great blue transmission. What is the advantage of using the red + IR bandpass rather than the BG3 Blue + IR filter other sites recommend and what would the small threadless filter cost me? Also, I understand why you would want to subtract the IR leakage in the red channel but is 40% a number you got from the reflectance calibration or just a rough estimate? I would be very interested in how you came up with that. Thanks for the post!

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

@Artichoke23 - What I have seen in testing is that a red filter will give you much greater contrast and sensitivity in crop vigor. A good example is corn. Stressed corn leaves will continue to absorb blue while reflecting much more red than healthy leaves. Ned Horning has a good note showing why red filters should work well with NDVI(https://publiclab.org/notes/nedhorning/11-01-2013/why-a-red-filter-should-work-well-for-ndvi) I'm not certain on the cost of specific filters but MidOpt can do custom stuff too(http://midopt.com/filters/db660850/) The 40% is what worked for specific settings on that Canon camera. I confirmed the NDVI of specific trees then adjusted the formula until I got the same value. The percentage of NIR in the blue channel to subtract from the red goes up dramatically if the images are not calibrated with reflectance targets.