New Audio Tool Identifies Bees Without Lethal Traps

Deep learning software now tracks honeybee and bumble bee activity through flight and vibration sounds with 97% accuracy.

Updated on Oct. 28, 2026 in Environmental

A parabolic microphone on a tripod nestled among wild sunflowers and clover in a sunlit meadow. — Researchers have introduced a non-invasive acoustic monitoring tool that tracks honeybee and bumble bee populations through flight and vibration signatures. AI Illustration. Upload story photo >

Live Poll

Should scientists prioritize using AI monitoring tools over traditional physical data collection methods?

Researchers have developed new software programs that use deep learning to identify honeybees and bumble bees by their acoustic signatures. The system, which operates within Raven Pro software, enables non-invasive monitoring of pollinator activity in difficult-to-reach areas like tree canopies.

Why it matters

Traditional pollinator survey methods often rely on soapy traps that kill the insects being tracked, limiting population monitoring efforts. This acoustic approach offers a non-lethal alternative for ecological research, potentially safeguarding the insects responsible for $200 billion in global services.

The system achieves 93% to 97% accuracy for reported detections when using a 0.9 confidence threshold. Each recording device is limited to an effective detection range of approximately 20 inches.

The players

UMass Amherst

A public research university that led the study on non-invasive pollinator acoustic monitoring.

Cornell Lab of Ornithology

The institution that developed Raven Pro, the sound-analysis software used to house the new detection programs.

The details

The team trained a deep learning network—a computational model that learns patterns from massive data sets—using 594 acoustic recordings. By analyzing the flight and flower-vibration sounds of bees, the programs isolate specific behavioral signatures. Researchers verified these automated outputs by manually auditing the findings against field counts performed by trained human observers.

Researchers collected 594 audio recordings at 43 sites between 2021 and 2026.
Figueroa demonstrated the effectiveness of deep learning for identifying bee sounds in 2024.

The Tech Race

This development integrates pollinator monitoring into the established bioacoustic research ecosystem managed by the Cornell Lab of Ornithology. By adapting Raven Pro for insect identification, the team moves beyond traditional avian sound analysis toward comprehensive multi-species ecosystem monitoring.

This software currently serves as an academic tool for researchers and ecological monitors to track bee populations without lethal sampling. Future updates are expected to expand the system's accuracy and geographic range beyond the northeastern United States.

The takeaway

Researchers are moving toward non-lethal acoustic monitoring to protect global pollinator health. Watch for future validation studies that test if these acoustic signatures remain distinct in regions with different ambient noise and biodiversity.