To be honest, I am skeptical. Here are my reasons:
By the time the mites can be seen on the backs of the bees in order to be able to count them, the infection level would be so high it is not worth bothering to count them.
Do not quote these numbers as specification as I have read so much literature, observing that the numbers vary between articles and am fuzzy as to which numbers come from which articles, nor verified any validity of them. That said, if nothing else one can bring them all together and get a ball-parking range of such values which should help get you thinking pragmatically about this.
- 95% of mites on the bees (phoretic or parasitic is current debate) will be on the underside of the abdomen. Your eyes, your phone, a camera, the AI app will not see those when looking at a frame of bees.
- depending on the time of year; typically 15% - 20% of the mites will be on the bees. The rest, 80%-85%, of the mites in a hive are hidden beyond view in the combs under the brood caps.
- For one mite to be seen on the back of a bee, there are likely another 2 to 3 more mites on the underside of that same bee which cannot be seen.
...... doing really simple math on those %ages, the mites that can be seen on the backs of the bees by the eye or a camera being analyzed by AI app represents less than 0.75% of the mites that are present in the hive. This is why I say that by the time you are seeing mites on the backs of the bees, the situation is so far advanced and dire that counting them is pointless. The infestation is so high that the bee and the colony are already dead. They just do not know it yet.
I appreciate the thought, the concept, and the project as being interesting from a R&D development exercise perspective. However, to be pragmatic about it I fail to see any practical use the app as proposed will provide to the beekeeper. Certainly agreed, varroa is seriously a bad thing and we all really want more tools to deal with them. But you are not going to discover nor fix a varroa problem looking at a picture of bees.
As myself being an innovator, I know it takes at least 10 bad ideas to discover 1 good idea. To be frank, the varroa counting app by phone camera and AI idea needs to go into the bad idea box. Not because of technical feasibility, but because there is no practical usefulness of it after one considers the points listed above.
All that said. Please do not be discouraged by my comments. Come up with at least 9 more ideas to discover a good one worth pursuing ... For example; instead of looking for varroa such an app could be developed and used to detect other visually apparent ailments such as; deformed wing virus in static photos, acute/chronic bee paralysis virus in short videos of quivering bees on the frame or the tell-tale hairless black shiny bees in still photos, sneaky hive beetles, worms, excessive dysentry on the frames, and/or brood indications such as AFB, chalk brood, sac brood, brood pattern quality and density, ... other ...
Therefore, my recommendation is to dump the varroa counting idea and instead steer this project towards something useful by focusing on those many other colony ailments that have clear indications which are easy to identify upon seeing them but which are too often obscured from the untrained/inexperienced eye looking at a frame full of bees and their stuffs.
Hope that helps ... In some way.