Posted on June 21, 2013
Today I worked on trying to load the slope data into eCognition and finally managed to bring it in with Briton’s help. After that I finally started to do some classifications using the slope.
I noticed that the objects classified as ferns were at the back of the valley and there were more on the left side of the valley or the side that does not have Atlantis, which was something I noticed when I was in Ka’a’awa Valley two days ago. So I knew that in general, using the slope worked pretty well, except for the fact that each object was so big that it included trees, ferns, and grass. Also, the program classified a road as ferns as well which was weird. So then, I tried to decrease the size of each object when the World View 2 imagery was segmented by changing the scale parameter and the compactness value. By changing the compactness value to 0.9 instead of the default 0.5 and having the scale parameter 100, each object seemed to be more accurate. I found that except for 2 objects everything was pretty much at the locations where I logged the GPS points or at locations I expected to see ferns because of what I remember about the unreachable ferns when I was at the valley a couple of days ago.
I will probably work on decreasing the scale parameter by 10s and see what fits better. I will also try to use the actual values of the pixel in WV2 imagery to eliminate the misclassified objects but I am still not sure about how to grab the couple of ferns that have not been classified as ferns at any point during my classification process. However, after managing to get the slope raster into eCognition, I am finally starting to enjoy using eCognition. Its ability to classify raster imagery is amazing.
I don’t think I need to go out to the field ever again since the number of fern patches in the valley is very limited and I have no reason to visit the site again since I remember what they looked like from close up as well as how they looked from the road on the opposite side of the valley.