Accuracy of YOLOv7-Tiny Varroa destructor detection using real mite data

Determine the true detection performance (precision, recall, and F1-score) of the IntelliBeeHive YOLOv7-Tiny "Mite" class detector for Varroa destructor by evaluating it on real, annotated occurrences of mites on honey bees rather than placeholder beads, to provide an accurate representation of the model’s effectiveness under realistic conditions.

Background

The IntelliBeeHive system trains a YOLOv7-Tiny model to detect pollen and Varroa destructor mites, but due to limited access to real mites, the authors used opaque red beads (1.5 mm) as placeholders when creating the training data for the "Mite" class.

As a result, the paper does not report validated accuracy metrics for mite detection under realistic conditions. The authors explicitly state the inability to accurately represent the mite detector’s performance because of the limitation in real mite data, leaving the true accuracy on genuine mite occurrences unresolved.