Automated, non-invasive Varroa mite detection by vibrational measurements of gait combined with machine learning.

Little is known about mite gait, but it has been suggested that there could be greater variation in locomotory styles for arachnids than insects. The Varroa destructor mite is a devastating ectoparasite of the honeybee. We aim to automatically detect Varroa-specific signals in long-term vibrational recordings of honeybee hives and additionally provide the first quantification and characterisation of Varroa gait through the analysis of its unique vibrational trace. These vibrations are used as part of a novel approach to achieve remote, non-invasive Varroa monitoring in honeybee colonies, requiring discrimination between mite and honeybee signals. We measure the vibrations occurring in samples of freshly collected capped brood-comb, and through combined critical listening and video recordings we build a training database for discrimination and classification purposes. In searching for a suitable vibrational feature, we demonstrate the outstanding value of two-dimensional-Fourier-transforms in invertebrate vibration analysis. Discrimination was less reliable when testing datasets comprising of Varroa within capped brood-cells, where Varroa induced signals are weaker than those produced on the cell surface. We here advance knowledge of Varroa vibration and locomotion, whilst expanding upon the remote detection strategies available for its control.