Hi, my name is Vera, and I’m in the final year of my biology bachelor’s degree. This past semester has been an exciting and challenging one, pushing me to step outside my comfort zone and dive into new ways of learning. One of the most unique experiences was through a project that let me explore pollinators, not through traditional fieldwork, but by working with artificial intelligence. Annotating images of pollinators might sound like a strange way to study biology, but it turned out to be a fascinating journey into how technology is advancing in ecological research.

Image from the annotation process using CVAT, with a visibly red bounding box drawn around a butterfly, annotated in the order Lepidoptera.

Using the Computer Vision Annotation Tool (CVAT; app.cvat.ai), my task was to identify and label pollinators in images captured through camera surveillance, covering entire flower life cycles. This involved drawing bounding boxes around the pollinators and classifying them into broad taxonomic categories, ranging from orders to species. Only identifiable adult-stage pollinators, belonging to defined groups, visiting the focal flower or hovering nearby were annotated.

One of the main challenges I encountered was the diversity of pollinators. Different taxa often appear similar, especially in blurry images or when captured in action. Fortunately, some of the taxa were broad, covering several species. Additionally, there was an ‘insect’ category for cases where a pollinator could not be classified into one of the specific taxa.

Image from CVAT: Is it a bee or a wasp? Such distinctions can be tricky during the annotation process.

The annotated data serves as training material for AI tracking systems, helping them learn to recognize pollinators in large datasets. By automating this process, researchers can track pollinator activity more efficiently, gaining insights into their presence and behaviors.

You might wonder, why is annotating images for AI so important? The answer lies in its potential to scale up the monitoring of pollinators across vast landscapes and time periods – something that would be difficult, if not impossible, to achieve through manual methods alone. By training an AI to identify pollinators, researchers can track their populations, behavior, and activity patterns over time, at a speed and scale that would be unmanageable through traditional field studies.

This kind of monitoring is especially important as pollinators face increasing pressures from climate change, habitat loss, and pesticide use. Understanding when and where pollinators are most active can help scientists and conservationists plan better strategies to protect them and the plants they pollinate.

While my experience didn’t involve fieldwork, it was no less rewarding. Annotating images to train AI for pollinator research was a fascinating process that allowed me to contribute to the broader scientific effort of understanding and conserving pollinators. As AI becomes a more integral part of environmental research, I’m excited to see how these technologies can help us learn more about pollinators and their behaviors that are so vital to the health of our ecosystems. As a biology student with a special interest in smaller creatures, this project provided me with valuable insight into the fascinating world of pollinator-plant interactions.