Anyone interested in helping with an opensource counting and detection app using python and Tensorflow or DLIB? Thinking about building a free opensource counting application, Just need the datasets (Imagery for training) to do it with. Wrote an application last year with no data or imagery using just colors (HSV and inrange functions with some other magic) no template matching or anything and got a really successful count (About 90% Accuracy) on fields with no cover crops, With imagery and training data I should be able to handle cover crops and all plus improve accuracy and do some weed detection also.
Hey brizey, I had discovered your post through a homework assignment I had to perform for my drone class and I found your post really interesting I had to come post here.
I am not familiar with TensorFlow or DLIB, but I will look it up nonetheless after this. But, I couldn't help but add maybe (if this can be implemented) a classifier/imagery training where the altitude of the drone and the area the specific crop type take up, and (now that I think about it) how the soil is around the crop type dark, light, etc., in terms of the x-y plane.
I am not too familiar with crops, these are some filters you can use, in my opinion and what i think is possible for detection. Furthermore, the altitude and area an crop type takes up in the x-y plane is probably a bit more realistic. Farmers can have the drone fly lower or higher to get a better detection of what kind of crop it is, especially for the weeds as you describe, the weeds being a lot smaller can be detected at a specific altitude range & area coverage maybe the actual crops are not detected. Possibly being more distinguishable of crops, weeds, even foreign animals or insects one day when farmers may run your app in a more autonomous application as the farmer drinks his coffee while watching through his drones onboard camera.
Let me know if that makes sense, I can reexplain in less words and more clearly.
Best,
Sanjit