Public Lab Research note


Trapa detection

by cfastie | August 02, 2013 15:56 02 Aug 15:56 | #8958 | #8958

Image above: Balloon aerial photo of aquatic plants on Lake Warner with patches of water chestnut circled. Click image to enlarge.

A plan is emerging to try to identify water chestnut (Trapa natans) in the balloon mapping photos we took a couple of weeks ago. It's easy to see water chestnut in many of the photos, and with a little practice one can mark individual rosettes quite quickly. But the hope is that after many hours of diligent effort by expert remote sensing professionals, a computer will be able to do this work without much help from us. In the meantime, I have tried the manual method of staring at a few of the photos and marking the patches I think are water chestnut. Don't show this note to the computers, because that would be cheating.

WarnerLkVis7075Trapa2500.jpg

One of our higher altitude photos of Lake Warner. All four of the plastic lawn furniture ground control points (red squares) are included. The yellow rectangles are the areas of the two photos below. White circles are patches of vegetation which I think are water chestnut. Right click, then open in new tab, then Ctrl+ to enlarge (but the site won't show you the full resolution photo!).

.

Most patches, and even individual plants or rosettes of water chestnut can be distinguished in the lower elevation photos, but smaller patches are not easy to see in the higher photos. Water chestnut has a blue green color that distinguishes it from most of the other vegetation in this lake. The leaves of water chestnut are smaller than those of most of the native plants here, so the texture of patches of water chestnut is often distinctive. These photos have been downsampled to about 70% of the size of the originals, and the "levels" tool in Photoshop has been used to increase dynamic range. Right click, then open in new tab, then Ctrl+ to enlarge (but the site won't show you the full resolution photo!).

.

WarnerLkVis6911Trapa2500.jpg

The water chestnut patches in this photo are very hard to distinguish in the higher altitude photo above. Right click, then open in new tab, then Ctrl+ to enlarge (but the site won't show you the full resolution photo!).

.

WarnerLkVis7177Trapa2500.jpg

The water chestnut patches (probably individual plants) in this photo are very hard to distinguish in the higher altitude photo above. Right click, then open in new tab, then Ctrl+ to enlarge (but the site won't show you the full resolution photo!).


8 Comments

Hello 10k Diamond Jubilee Admirals' Club. This is my 100th research note!

Reply to this comment...


chris fastie for president. i think tallow looks like broccoli from up high. it would be interesting to develop a color signature, or some kind of machine-readable "pattern" for different invasive species.

The advantage of Color Infrared is that different plants have different phenologies, so that there may be optimal times of year to take pictures than can discern plants like trees from one another. at the class I took at USGS, the pearl of wisdom was that fall photography was always the best--half because we were talking about wetlands, and in our corner of the world, fall is a time of low water, too. but the changing season allows for photo interpreters to distinguish between species in an aerial photograph.

Reply to this comment...


Yes, I noticed that in your note you mentioned that tallow turns color in the fall and nothing else does. So you won't even need IR to distinguish it. That will make life much easier. At other times of year, or to distinguish the other plants from one another, IR can be helpful. Around here you need to time the fall photos carefully because color happens fast. I tried to identify some trees in the snapshots of this late October gigapan: http://www.gigapan.com/gigapans/35627. Every week is completely different.

Reply to this comment...


pretty nice results! it seems like the structural information is more telling than the spectral. did you try to calculate e.g. spatially resolved fractal dimensions? what i'm playing around with for similar structures (algal patches) is spectral coherence. if you have a high-res pic available for download i can give it a try...

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

Reply to this comment...


philippg, Ned Horning has been working on these images, and has tried some segmentation approaches, but maybe nothing fractal or with spectral coherence. It would be good to hear if you can learn anything using those techniques. Ned has put two 4-band tiffs here. These have an NIR band added to the RGB photo.

Some of the other clearer images can be downloaded as follows: In this Google Doc is a list of 43 of the balloon photos with some metadata including the file name of the matching NIR photo. The twelve largest VIS photos and their NIR matches can be downloaded as in: http://fastie.net/Bin/12WarnerVIS/WarnerLkVis6911.JPG or its NIR match: http://fastie.net/Bin/12WarnerNIR/WarnerLkNIR2051.JPG

If the Google doc is sorted with largest (Size) at the top, the first twelve pairs can be downloaded by substituting the 4 digit file number in the above URLs (unique directories for NIR and VIS). These are 10 MP photos. The ones with the largest file sizes tend to be the sharpest.

Let us know what you find.

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

Reply to this comment...


Chris, some questions for you:

1) Is there a full resolution version of WarnerLkVis7075.jpg and WarnerLkVis7136.jpg somewhere for me to download? It's not available at http://fastie.net/Bin/12WarnerVIS/. That large patch of Trapa in the large circle at bottom center of 7075 is useful for grabbing Trapa color info, and I want to include NIR, but would like to avoid potential mismatch pitfalls in compressing/downsampling the NIR photo myself.

2) Are your NIR images as they came off the camera? (if so, did you set the NIR camera to record as B&W? Often it's possible to get red edge spectral info if you record NIR as color, depending on the characteristics of the Bayer array on the your camera)

Thanks!

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

Reply to this comment...


Tom,

Both WarnerLkVis7075.JPG and WarnerLkVis7136.JPG are there (case sensitive). Their NIR matches (2214 and 2275) are there too.

The NIR images have been passed through Adobe Lightroom, but only to add metadata -- the pixel data should not have been altered. The NIR camera was custom white balanced but was capturing color photos. So all three histograms (RGB) are completely overlapping and the photo appears as grayscale. The camera has a Wratten 87 IR filter, so only NIR above 735 nm made it to the sensor, but the Bayer filters had already done their thing.

Chris

Reply to this comment...


I just posted a research note with my little background study of the usefulness of color for Trapa classification; Here's the link.

Reply to this comment...


Login to comment.