First prototype of that is ready, though I'm going to need so many more pictures before it's useful. It's actually very similar to a Microsoft sample in which they look at pictures and determine if something is cracked or not cracked.

https://github.com/dotnet/machinelearning-samples/tree/main/samples/csharp/getting-started/DeepLearningImageClassificationBinary

Still needs some adjusting and of course more pictures but it's getting there. Even with the limited dataset, it's getting some results because almost all rapid tests follow that specific physical cassette form which is going to help. Small circle and a larger rectangle with at least one line (control) though right now I'm trying to keep it simple and use images of only the RATs that have two lines max.

If you feel like helping out, send me images! Here are the rules:

  • Looking for fully or partially visible entire cassette in the picture. "Partially visible" means that fingers are holding the cassette, the entire cassette is not visible.

  • Either a positive result or negative, no blanks.

  • No bad tests (no control line).


Here are some negative tests, either fully or partially visible: