Challenge 2:
Inspiration
What's the best way to skin a cat? Not sure, but there are several ways to find tillable land in an aerial image of arable land. We wanted to explore our options and find a clever, quick way of solving this problem.
What it does
tillable uses a stack of filtering, thresholding, and morphological manipulations to find zones in an aerial image that would most likely be arable land.
How we built it
Trial and error.
Challenges we ran into
There are a lot of ways to do this problem, and we spent maybe 4-5 hours exploring other options before we settled on a hybrid SLIC/K-means based dilated filtering method.
Accomplishments that we're proud of
We wrote 4 very fancy end-to-end implementations of different algorithms that didn't work.
What we learned
Sometimes the simple solution is the best solution.
What's next for till.able
Calculation of arable land is a trivial addition to our implementation. The real next step is implementation of some kind of neural net, probably a ConvNet for classification of non-arable land areas, such as recognizing houses, barns, streets, rivers, etc. a la https://project.inria.fr/aerialimagelabeling/
Log in or sign up for Devpost to join the conversation.