An AI-Based Open-Source Software for <i>Varroa</i> Mite Fall Analysis in Honeybee Colonies

Infestation by <i>Varroa destructor</i> is responsible for high mortality rates in <i>Apis mellifera</i> colonies worldwide. This study was designed to develop and test under field conditions a new free software (VarroDetector) based on a deep learning approach for the automated detection and counting of <i>Varroa</i> mites using smartphone images of sticky boards collected in honeybee colonies. A total of 204 sheets were collected, divided into four frames using green strings, and photographed under controlled lighting conditions with different smartphone models at a minimum resolution of 48 megapixels. The <i>Varroa</i> detection algorithm comprises two main steps: First, the region of interest where <i>Varroa</i> mites must be counted is established. From there, a one-stage detector is used, namely YOLO v11 Nano. A final verification was conducted counting the number of <i>Varroa</i> mites present on new sticky sheets both manually through visual inspection and using the VarroDetector software and comparing these measurements with the actual number of mites present on the sheet (control). The results obtained with the VarroDetector software were highly correlated with the control (R² = 0.98 to 0.99, depending on the smartphone camera used), even when using a smartphone for which the software was not previously trained. When <i>Varroa</i> mite numbers were higher than 50 per sheet, the results of VarroDetector were more reliable than those obtained with visual inspection performed by trained operators, while the processing time was significantly reduced. It is concluded that the VarroDetector software Version 1.0 (v. 1.0) is a reliable and efficient tool for the automated detection and counting of <i>Varroa</i> mites present on sticky boards collected in honeybee colonies.