Automated Detection and Monitoring of Ground-Nesting Bee Nests Using Drone Imagery and Deep Learning

Area of Science:

• Ecology
• Conservation Biology
• Computer Science

Background:

• Ground-nesting bees are crucial for pollination and soil health, yet many species face threats.
• Limited knowledge of nesting habitat requirements hinders conservation due to difficulties in locating nests.
• Efficient monitoring methods for ground-nesting bee populations are lacking.

Purpose of the Study:

• To evaluate the feasibility of using drone-based imagery and deep learning for automated detection of ground-nesting bee nests (tumuli).
• To distinguish bee nest tumuli from other soil surface features, such as earthworm casts.
• To provide a scalable system for monitoring bee nesting sites to support conservation.

Main Methods:

• Acquisition of high-resolution aerial imagery using drones over a 120 m² area.
• Application of deep learning models to detect and classify soil mounds (tumuli) as bee nests.
• Validation of the model against ground-truthed data, distinguishing tumuli from earthworm casts.

Main Results:

• The deep learning model achieved a high F1 score of 0.90 (precision: 0.89, recall: 0.91) in detecting bee nest tumuli.
• The system accurately distinguished bee nest tumuli from earthworm casts.
• Misclassifications were primarily due to atypical tumuli shapes or overlapping mounds.

Conclusions:

• Drone-based imaging combined with deep learning shows significant potential for efficient monitoring of ground-nesting bees.
• This automated approach can provide crucial data for understanding bee nesting biology and supporting conservation strategies.
• Future research should explore broader applicability across diverse habitats and species, and investigate alternative methods like image segmentation.