Combining Camera Trapping and Deep Learning for Labor-Saving Monitoring of Perching <i>Sympetrum</i> Dragonflies （Odonata: Libellulidae）

A camera-trapping system developed for photographing perching Sympetrum dragonflies using passive photosensors is expected to support the biodiversity monitoring and evaluation of rice paddy fields. However, considerable time and labor was required in order to visually check the massive number of captured photographs to determine whether the dragonflies were correctly captured in them. In this study, automatic image classification using a deep-learning artificial-intelligence （AI）model was applied to the photographs captured by camera traps set in rice paddy fields in eastern Fukushima prefecture. The results show that the AI model can accurately detect and classify photographs containing a Sympetrum dragonfly, and that the estimated detection frequency based on the AI image classification is associated with the relative density of Sympetrum dragonflies directly observed in the fields. This suggests that automation by combining camera trapping and AI image classification is promising for monitoring Sympetrum dragonflies in the study area.