Researchers at Oregon State University have created a low-cost, semi-automated, AI-enabled monitoring approach that uses remote cameras to survey bumblebees and potentially other insects. The work targets a major challenge in conservation biology: obtaining enough reliable population data across large landscapes without relying on expensive or harmful sampling methods.
Insect declines have increased urgency around pollinator monitoring, and several bumblebee species are under consideration for listing under the Endangered Species Act. Yet traditional survey techniques are often constrained by cost, labor, and biological impact. Traps can be lethal, while direct observation with nets is time-intensive and difficult to scale.
The new system builds on earlier findings that trail cameras can capture insect imagery without direct interference. Previous camera-based studies focused largely on nocturnal insects that can be attracted with light, leaving day-active (diurnal) species less studied in camera-lure settings.
To address this gap, the team designed an inexpensive camera station assembled from commercially available components and tested it during bloom in a red clover seed field at the Hyslop Field Research Laboratory in Corvallis. In parallel, researchers collected conventional survey data using hand-netting and blue vane traps to provide an independent benchmark.
Image-based identification relied on two custom deep learning models. Both models processed the captured photographs, but the researchers found that “tiled” models—dividing images into smaller regions before classification—outperformed approaches that evaluated entire images at once.
Attracting bees to the cameras proved to be another key design variable. The researchers tested multiple visual patterns placed behind the camera stations and report that bullseye patterns generated significantly more bumblebee visits than uniform backgrounds.
Across their field deployment, the camera system recorded six bumblebee species, closely matching the diversity observed through netting and traps. This alignment suggests the approach can reproduce conventional survey results while avoiding lethal collection.
The broader implication is scalability: automated, wide-area monitoring could produce richer, higher-volume datasets needed for habitat and protection decisions. With data gaps often complicating listing determinations and critical habitat designations, the team argues that camera-plus-AI pipelines could make those processes more precise.
The findings were published in Remote Sensing in Ecology and Conservation. The research team also emphasizes potential agricultural relevance, since many crops depend on insect pollination and may benefit from improved monitoring of pollinator health and movement trends.
Subject of Research: Bumblebee monitoring using remote cameras and deep learning
Article Title: Low-cost AI-driven remote camera method for surveying bumblebees (exact title not provided)
News Publication Date: Not provided
Web References: https://statusofinsects.github.io/index.html
References: Remote Sensing in Ecology and Conservation (journal name provided)
Image Credits: August Jackson, Oregon State University
Keywords: bumblebees, pollinator conservation, remote sensing, trail cameras, deep learning, species identification, habitat protection, agricultural pollination
