AMI: artificial intelligence in an insect trap
Maxim Larrivée, director of the Insectarium de Montréal, and myself, an entomological assistant at the museum, traveled to Panama in January 2023 to join an international research group developing an automated system for monitoring nocturnal insects. This device is called AMI, for Autonomous Monitoring of Insects, and can record the presence of thousands of insects every night. The idea for this state-of-the-art system was born in the heads of Kim Bjerge and Thomas Toke Høye, two researchers from Aarhus University in Denmark. The beautiful thing about AMI is that it’s 100 percent independent thanks to its battery and its solar panels and that, unlike humans, it never gets tired! Night after night, AMI activates automatically. Its UV light turns on and attracts insects to a white surface. The movement of these insects on the surface triggers a high-definition camera, which immortalizes their portraits. This is a non-lethal trapping method, since the insects are not captured. They can leave at any time. The thousands of photos taken are recorded in the AMI’s “brain”: a hard drive connected to a microcomputer. A sequence of three artificial intelligence (AI) algorithms then analyzes each pixel to count and identify the insects photographed.
A better understanding of the decline in insect populations
More and more studies indicate that insect populations on a worldwide scale are declining in biodiversity and in biomass. Several hypotheses might explain the causes of this decline, but a lack of data makes it difficult to determine the main factors. Right now it’s complicated to do standardized monitoring of insect populations, since trap installation and collection in traditional inventory methods are labor intensive, as are the preparation and identification of specimens captured. The number of sites sampled and/or the sampling frequency are therefore low, and don’t allow for an authentic portrait of the situation of the thousands of insect species in an ecosystem. Those ecosystems are also very often remote and difficult to get to. On top of that, traditional traps are almost all lethal: insects are killed before being counted and identified. It was the lack of data that moved scientists to join forces in order to develop an automated non-lethal insect monitoring trap.
Testing the system in difficult conditions
The main goal of the expedition to Panama was to test the system in the tropical forest. That ecosystem presents challenges for AMI: humidity, heat, difficult access, solar panels rendered ineffective by the canopy, considerable falling of branches and leaves, and so on. Nevertheless, the tropical forest is one of the ecosystems where AMI deployment could be the most profitable, since the insect community there is extremely diverse but surprisingly very little known. That lack of knowledge poses another major challenge for AMI: its AI has trouble identifying the insects photographed. The database that the AI depends on for its analyses isn’t well stocked enough. We therefore made our way to the Smithsonian Tropical Research Institute on the Barro Colorado Island to run some workshops and carry out four nights of tests. In studying the tons of photos taken by AMI on this short visit, we estimate that we identified more than 500 species of moths! The most incredible thing? About 125 of them would be new species. Proof of the relevance of a tool like AMI for a better understanding of insects and learning to protect them.
- Bjerge, K., Nielsen, J. B., Sepstrup, M. V., Helsing-Nielsen, F., & Høye, T. T. (2021). An automated light trap to monitor moths (Lepidoptera) using computer vision-based tracking and deep learning. Sensors, 21(2), 343.
- An Automated Light Trap to Monitor Moths (Lepidoptera) Using Computer Vision-Based Tracking and Deep Learning