This week saw a series of thunderstorms pass over eastern Oregon. They get some fantastically powerful lightning events this time of year. I was pointed to this wonderful time lapse video of storms.
I was entranced by the beauty of the turbulent flow evident in the clouds. Pay particular note to the rippling waves in the early views of the sunset (~6:42); amazing.
Watching the video brought to mind being in a real drencher I experienced in Redmond, OR. I was on my way to explore some lava tubes. While driving between Bend and Redmond we were treated to lightning flashes all around us. As we neared town, fat rain drops fell in such volume that I couldn’t see the road so I pulled over to let the rain pass. The nearby lightning strikes and the accompanying thunder were awesome. I was much more concerned with someone finding the side of my truck with their bumper before we left the highway than being struck by lightning.
I don’t live in an area prone to wild thunderstorms or with wide horizons. Through most of the year, low coastal overcast sits atop the local hills shutting off any view beyond their slopes. However, at this time of year, I can climb the small hills in the neighborhood to get a pretty good view to the coast range on our west. And due to prevailing onshore winds, that is where most of our clouds come from. I’ll have to pay more attention to how much turbulent flow I can see in our clouds. The underside of the overcast tends to be uniformly smooth. And thinking about this brings me back to my current hair shirt.
I’m working on an interesting project that my mind just won’t leave; an electric transmission network stability analysis. In one part, I characterize the health of vegetation along the conductors. I’ve been doing this for over ten years, but this time the results are not quite as good as I would like. I think they will improve with better image texture characterization. Given advances in medical imaging, what I’ve been doing up to now seems pretty simplistic. Watching these wonderfully textured clouds has sparked some ideas.
One impediment to successful texture description is there is no formal mathematical definition of texture; meaning texture is situationally dependent. I’ve been using measures of smoothness, coarseness, and regularity in five color bands to describe texture. While this gives a good description of the changes in color it yields little on the arrangement of the colors or their spectral distribution. I think that at least in describing clouds this would be important; I hope the same is true in trees.
Our eyes, and my current algorithm, pick out color and shading; but I want to discriminate between the textures of regions containing wispy clouds in front of laminar clouds from turbulent clouds. And since it is only a means to and end, I must calculate it fairly quickly. These types of algorithms are used extensively in medical image analysis—particularly MRI, but some, if not most, are painfully slow. And I don’t think I need something that complex.
Watching the video I see that the arrangement of the pixels might help discriminate more classes, so I’ve been reading a few papers on finding tumors in MRI images.
Here is an example Aerial photograph image:
In this true color image you see several species of hardwoods and conifers mixed together in a mixed forest along a road in western Oregon. The dead tree (the red one) is near the corner of a treatment boundary. The boundary extends roughly from the dead tree to the lower right and from the dead tree straight down. The trees to the lower right are nearly all Douglas-fir while the rest of the forest is a mix of big-leaf maple, black cottonwood, Douglas-fir and western hemlock. There is a recently planted clear cut on the very left of the image.
Using only true color data I can pick out the dead tree near the center of the image. Another tree is starting to lose its leaves about half way to the right side on the treatment boundary; it’s marginally detected. Adding a bit of infra-red to the true color data I can separate the conifers from the hardwoods. The hard part to do automatically is to discern the leafy blobs from the spiky blobs.
The leafy blobs are the maples, having larger leaves they are more discernable as distinct objects in the image. It would be very wonderful to create a texture definition that will separate the two species.
I have a few gained a few leads from reading the MRI literature, but I’m being careful to limit the time and effort it takes to implement them.
Here is an example MRI image:
I don’t know anything about the medical implications of what I’m seeing here but I’m very impressed with the level of detail.
This image is reconstructed from the magnetic variations detected by the instrument. No light was used in the creation of this picture.
My eye picks out tonal variations, like the color summaries I use already, but there are striations visible in the muscle to the lower right. That’s spatial and might show a difference in the spiky trees.
Once I identify the dead and diseased trees then I have to work some magic to guess how likely they are to fall and if they fell how likely are they to hit the wires. This is a teeny little peek into my working world inspired by watching thunderstorms.
Too much, well you can calm down watching meteors over the desert.