Deep Learning-Based Monitoring of European Hamster Activity

In order to ensure the effective conservation of the critically endangered European hamster (Cricetus cricetus), there is a necessity for the implementation of targeted conservation measures and reliable monitoring methods. This study explores the potential of employing Deep Learning (DL) to assist with camera trap monitoring for the purpose of tracking hamster activity. To this end, an object detection model (YOLO) was trained to efficiently analyze large volumes of video data from summer 2023 with high reliability. The model achieved a weighted average F1-score of 0.93 and an accuracy of 0.93 for the detection of European hamsters, effectively differentiating them from other species. A comparison between DL-based and human evaluations confirmed that DL can reliably depict hamster activity patterns. The findings of this study suggest that European hamsters exhibit peak activity levels at dusk, with the highest peak in activity occurring around sunset. In contrast, activity levels were lowest around midday. Autocorrelation analysis revealed a biphasic activity pattern, with a secondary peak occurring approximately before sunrise. This study underscores the potential of employing DL for long-term conservation efforts and its applicability in assessing the success of reintroduction programs.