Public Lab Research note


by cfastie | August 18, 2013 03:52 18 Aug 03:52 | #9031 | #9031

It's a good year for tomatoes, and a big year for hornworms. These caterpillars fatten on tomato leaves so they can become huge Carolina sphinx moths (Manduca sexta). As fantastic as they are, they are very hard to find. Even when I see lots of tomato leaves missing, I often can't find the culprit because they are so well camouflaged. In a fleeting moment during which I considered that I was smarter than a hornworm, I decided to use science to overcome their defenses. Although they are the same green color as tomato leaves, everybody knows that caterpillars don't absorb red or blue light for photosynthesis, so they should look distinct to an Infragram camera. Or so I thought.


Tobacco hornworm on my tomato plant.

I took three different kinds of infrablue photos of the hornworm above:

  1. Canon A810, Rosco #74 filter, white balanced on blue paper under blue sky in the shade.
  2. Canon G11, professional Superblue filter (Schott BG3), white balanced on blue cloth under blue sky in the shade.
  3. Canon G11, professional Superblue filter (Schott BG3), white balanced on 18% gray card in the sun.

The false color infrared image (NBG) was made in Photoshop the same way for each photo. The normalized difference vegetation index (NDVI) image was made in Ned's Fiji plugin the same way for each photo (no stretching).


Infragrams do a poor job distinguishing tobacco hornworms from tomato foliage. It seems that the caterpillars reflect the same proportion of blue and infrared light as do tomato leaves, which was puzzling. So I Googled it. Sure enough, the larvae take ß-carotene from the leaves and move it to their skin. Tobacco hornworms fed a diet without leaves turn blue. I'm not sure that is why the larvae look like leaves to an Infragram camera, but apparently the ß-carotene in their skin is still absorbing blue light and reflecting near-infrared. Very clever larvae.

I did this Help out by offering feedback! Browse other activities for "multispectral-imaging"

People who did this (0)

None yet. Be the first to post one!


Time to begin scanning them with a spectrometer! Don't you think there ought to be more differentiation there?

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

Reply to this comment...

The lack of differentiation really surprised me. The fact that hornworms fed a leaf-free diet turn blue is consistent with the idea that ß-carotene in the skin is mimicking the Infragram signature of leaves. Carotenoid pigments look orange because they absorb blue but not red. Without the carotenoids, hornworm skin reflects more blue. Although my hornworms were apparently absorbing blue, like leaves, they should not be absorbing so much red (chlorophyll in leaves absorbs both blue and red) . So I hypothesize that a two camera infrared system that uses red light instead of blue for the visible band will distinguish hornworms from foliage. If someone would like to fund this research project, I will send you my PayPal info.

Reply to this comment...

Have you attempted to do dual-camera red/IR photos? Or even just a regular visible-light photograph of the same worm to get the red channel?

What do you do with the worms (i'm almost afraid to ask) when you find them?

Another thought -- if we ever get around to doing hyperspectral imaging with the smartphone specs, we could try to resolve a "heatmap" of the worms. I was thinking that something like Infragrammar would allow us to select specific wavelengths to generate use-specific images from a hyperspectral scan.


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

Reply to this comment...

I haven't wrapped my head around how to make a DIY hyperspectral image. I guess the idea is to make a photo of an object by capturing only a very narrow band of the wavelengths reflecting off the object. Then repeat that for each of many, many narrow bands. Then stack all the photos, so you have data for each spectral band for each pixel of the image. That is, each pixel is a spectrum instead of one color. That's a lot of data.

I tried to give Ned's Fiji plugin pairs of RGB and infrablue photos of hornworms to make NDVI from a red and NIR channel, but it would not align the handheld photos. If I find more hornworms I might try it with the cameras on a tripod. The last hornworms are gone now. They are good grilled with an Argentinian Malbec.

Reply to this comment...

... and I always run to grab my regular camera when I see a Sphinx moth flying in the yard!

Reply to this comment...

Well, a common way to do hyperspectral is to just collect a whole row of spectra (y dimension is your row of pixels, x dimension is wavelength) and then collect a stack of them for each column of light from the real world -- or more conveniently, a video of slowly moving past the scene.

In airplanes and satellites, the time dimension is usually just the motion of the camera itself across the landscape. To make my apple photo, I dragged it slowly across the frame on a black piece of paper while illuminating it VERY brightly with floodlights.

Reply to this comment...

Login to comment.