Automated land cover classification for conservation and habitat monitoring

I posted the results of my research into using a small UAV for land cover classification for habitat monitoring a couple of months ago.

I've created a short video as part of my research presentation that details the workflow used to analyse the image data and thought it may be of interest to the community. I'm more than happy to have a conversation about the project in the comments :)

Views: 851

Comment by Martin on May 8, 2015 at 6:10am

I read your previous post and you said you used imageJ for the falsecolor image. I think it's a much better idea to use GIS software like QGIS for the purpose. All the georeference data is preserved that way.

Could you talk about the automation part of your work. Are you using a learning algorithm for classification?


Moderator
Comment by Gary Mortimer on May 8, 2015 at 6:48am

Nice, very nice

Comment by Mark Williams on May 8, 2015 at 7:22am
Hi Martin, you're right, generating the NDVI image in the GIS software is better and that is what I'm doing now.

I'm using nearest neighbour classification in eCognition which is a learning algorithm of sorts.
Comment by Martin on May 9, 2015 at 1:57am

Hi Mark,

Could you go into some more detail on how you're using eCognition? Maybe some screenshots or a video.

Comment by Mark Williams on May 9, 2015 at 2:10am
Hi Martin, there's lots of YouTube videos about nearest neighbour classification in recognition. This one is on version 8 but should give you a good idea of how it works. https://youtu.be/vVQBiWWIAuM
Comment by Martin on May 9, 2015 at 6:05am

Thanks Mark, the person in that particular video does not seem too fond of the user interface though.

Comment by Adam Erickson on May 9, 2015 at 11:59am

These papers loosely describes eCognition's nearest neighbor supervised classification algorithm with fuzzy rules (link and link). It is based on the feauture space of the sample objects. eCognition provides a rich toolset for nearest neighbor classification that includes spatial autocorrelation and other contextual descriptors. In other words, it is principally object-based. There are an infinite number of ways one could mathematically describe what a nearest neighbor is, but here it refers to distance in the sense of multivariate space and either Euclidean or Mahalanobis distance, in what is likely a variation on the commonly used k-nearest neighbors (knn) algorithm.

Comment by Mark Williams on May 10, 2015 at 1:50am
Hey Adam. That's my understanding of it as well though in perhaps less florid terms. I'm still in the learning stage with a lot of these classification techniques but they're very interesting and finding the right technique and its use for the toolset is the challenge.
Comment by Mark Williams on May 11, 2015 at 7:08am
Martin, I'd agree. The UI can be awkward at tikes but you kind of get used to it.
Comment by Mark Williams on May 11, 2015 at 7:09am
*times

Comment

You need to be a member of DIY Drones to add comments!

Join DIY Drones

© 2019   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service