Public Lab Research note


Multispectral DeFelice

by cfastie | April 03, 2013 04:17 | 3,123 views | 4 comments | #6653 | 3,123 views | 4 comments | #6653 03 Apr 04:17

Read more: stable.publiclab.org/n/6653


Image above: NDVI image of part of the front yard at the LUMCON DeFelice Marine Center in Cocodrie, Louisiana.

Caterina has video footage of Scott describing the yellowish appearance of autumn Spartina marsh in false-color infrared aerial photos. She was looking for an aerial image to illustrate this, but I had never processed all the photos from the big kite flight around the DeFelice Marine Center in Cocodrie, Louisiana. This was the perfect excuse not to work on my taxes, so I fired up Ned’s ImageJ plugin and fed it the 650 photo pairs from the Public Lab infrared camera rig I flew on November 3, 2012. The plugin worked magically after referring to the manual to remind myself of a few crucial spells.

I tried several combinations of color channels for computing normalized difference vegetation index (NDVI) and settled on using the blue channel from the normal camera for the red band and the red channel from the IR camera for the infrared band. The color table that I used to display the NDVI values is one that Ned made to be friendly to those with red-green colorblindness. Dark green is the highest value which represents the greatest amount of photosynthetically active biomass. White is the least amount of plant biomass, and purple is non-photsynthetic stuff, like soil, water, wood, and dead plants.

The poster below has the normal true color image, the false color infrared image, and the NDVI image of part of the front yard at the LUMCON facility. The true color and false color images are 23-24 images stitched in Microsoft ICE, and the NDVI image is 8 images stitched in ICE and five more added in photoshop. ICE had a harder time finding control points in the NDVI images. [Update: Embeds from gigapan.com don't display anymore. See it here: http://www.gigapan.com/gigapans/126829]

These maps will have to be done again by hand in Photoshop to get the best georectification. The area in this scene is only about 20% of the area covered during the flight, so I have a lot of aligning to do.

I used the defaults in the plugin except for: Affine, nir(g+b)vis(g-b), jpeg, the channels for NDVI, and the lut.


4 Comments

thanks Chris!

Reply to this comment...


This is an incredible map -- the variability in color of different patches of vegetation makes me think it'd be easy to do a supervised classification here...

I am also testing comment notifications on the new site. But don't let that diminish my plaudits.

Reply to this comment...


Very nice Chris. I'd been wondering how those image pairs from Cocodrie were doing. Great to see this subset processed and stitched.

Reply to this comment...


Thanks Pat (It's like Deja Vu).

Jeff, I'm not sure I got a notification of your comment. But I might be confused.

There are two very similar plants that form the broad roundy patches in the lawn. We might not be able to distinguish them even if we knew which was which in order to train the classifier (which we don't). But a classifier should be able to distinguish those clonal, mat-forming plants from the grass (Bermuda grass?) growing on the artificial mounds of topsoil. But we already know where they are.

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

Reply to this comment...


Login to comment.