How did they locate these particular images? I would love to read something on that, or similar projects looking across satellite imagery for specific patterns.
Easiest way is going to be to downsample the images and then apply a pre-trained classifier that can ignore the fact these are sat images. You could probably turn them into 28x28 greyscale and then use a model trained on handwritten characters, like EMNIST.
Whatever approach you take, you'll probably be selecting the final set by hand, so it's just about building the candidate set in an efficient manner. Low absolute accuracy isn't really an issue as long as you end up with a managable set to review.
There simply doesn’t seem to be that many different images so I can imagine an even easier way than an image classifier.
As you work through your day as a satellite image person (I don’t know the job title, sorry) and see a shape that looks like a letter screenshot and save it. You’ll have an alphabet in a very short amount of time.
Given the repeats of letters I’m seeing I think that’s all they’ve done here.
Whatever approach you take, you'll probably be selecting the final set by hand, so it's just about building the candidate set in an efficient manner. Low absolute accuracy isn't really an issue as long as you end up with a managable set to review.