Check out our store for discounts and new items, & thanks for your support in 2017!

Public Lab Research note


  • 5

Plant health NDVI Red vs Blue filter

by Claytonb |

While testing Ned Horning's calibration plugin(Ned's plugin) I had the opportunity to compare some red and blue filters and their contrast and sensitivity in relation to NDVI values and plant health.

The filters I tested were the MidOpt DB 475/850, MidOpt DB 660/850, and Event 38 NGB filter. Each were placed internally into Canon S100's that had their internal factory IR filter removed. RAW images were captured from each camera with f/4, 1/1000, ISO 100 on all. They were each calibrated against 4 calibration targets with known reflective values using Ned's plugin. The NIR subtraction is still subjective as I do know yet know exactly how much NIR is in the visible band on these camera's. The leaves in the image were from a brush type plant in my front yard that I noticed had different stages of visible health. The red filter showed much better sensitivity to leaf condition.

I am curious if anyone has images they can share that show certain crops with better contrast using a blue filter vs red filter. So far there has been strong evidence that red is the go-to filter for NDVI. I would also like to know if anyone has information on the full spectral response of this Canon sensor.

Ned has provided sound logic on why red filters should perform better than blue( see this research note) when looking at plant vigor or health so I was curious why some still use blue filters for NDVI analysis and what exceptions may exist when a blue filter would be better for this application.

RGB Image

0207161357_HDR_(2).jpg

MidOpt DB475/850 VIS blue NIR red

475-850_zoom.jpg

Event38 NGB VIS blue NIR red

Event38_zoom.jpg

MidOpt DB660/850 VIS red NIR blue

660-850_zoom.jpg

Full NDVI image from MidOpt 660/850(red filter)

660-850_with_bar.jpg

Another RGB

Closeup_RGB.jpg

MidOpt 660/850 VIS red NIR blue

ndviImage_with_bar.jpg

MidOpt 475/850 VIS blue NIR red

ndviImage_with_bar.jpg



ndvi calibration infrared infragram wratten25a super-red red-vs-blue

response:9717


9 Comments

This is really interesting. I have read Ned's article too and it is clearly better contrast when using red filter versus blue.

I have just received my Canon s100 from Ebay and deciding on which red filter should I get. The one you used above (MidOpt DB660/850) looks like has some sort of noise compared to Event38's filter, do you know why?

What do you think about this red filter ? https://www.rosco.com/filters/SED.cfm?titleName=R26:%20Light%20Red&imageName=../images/filters/roscolux/26.jpg

These guys do also have some filter but I am not sure which one would fit.

Could you clarify a bit what do you mean by "VIS blue NIR red" and "VIS red NIR blue" ? They I understand it is that blue is not used at all in the red filter so why it is in the equation at all ?

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


@bortek -I am assuming that the noise is partly due to the filter thickness and light requirement. I believe the Event38 filter is 0.5mm thick and the MidOpt is 1mm. They are both thicker than the factory IR filter. I spent a lot of time adjusting the focus with both filters and got them improved. Here is an image with the MidOpt filter from about 350ft with the final focus. The edges are not great but they stitch ok. This is with f/4, exp 1/1600, ISO 400

IMG_0056.JPG

Are you looking for a filter for a single camera or dual camera NDVI? A 8.9X7.9 filter will fit nicely in the Canon S100. The channels correspond to the filter I'm using and the channels on that camera to calculate NDVI. So with the Red filter the Red channel(VIS red) is used for the visible band in NDVI equation and the Blue channel(NIR blue) now records NIR for the equation. The downside with single camera NDVI is that all channels record some level of NIR.

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


Here is a similar scene with the Event38 at about the same height. This is with f/4, exp 1/1250, ISO 100

IMG_0064.JPG


@Claytonb - I am planning to start with a single camera and do some initial testing and learning before I go over to two cameras.

Yes the focusing can be improved but it might not be so important as we are not planning in zooming so much.

This is funny because I dont really get the whole process in my mind yet. If we use a red filter then it will pass visible RED light and NIR. In your first picture under the comments the grass looks blue. How come the blue channel is registering NIR and why not for example a green channel registering it instead. Is the filter changing the wavelength of NIR to visible blue ?

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


@bortek -the blue in the NDVI image is related to the value of the color LUT I applied to the image during processing. I did that to help me to see where I have shadows. This is what the separate channels look like with the MidOpt 660/850. Because the filter only passes red and NIR the red channel on the camera records visible light and now the green and blue channels record mostly NIR. The blue channel has a much cleaner NIR signal. Try downloading the parking lot images above and run them through Infragram Sandbox http://infragram.org/sandbox/ using different channel combinations to see how the channels differ in NDVI calculation. Red Channel

Red_Channel.jpg

Green Channel

Green_Channel.jpg

Blue Channel

Blue_Channel.jpg


Thanks Clayton. I am on it now. :)


@Claytonb - I was just wondering if you could indicate what type of calibration targets you are using? I've been looking at targets produced by labsphere, which seem to be very high quality but haven't found any more.

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


@dannysmith -I have been using the Labsphere Permaflect coated targets http://labsphere.com/labsphere-products-solutions/materials-coatings-2/coatings-materials/permaflect-1/ their 2"X2" targets are in the $300 range and work great. I suggest contacting Charlie Langlois with Labsphere- clanglois@labsphere.com


thanks @Claytonb, they seem like the way to go


You must be logged in to comment.