Pest detection and trapping
Invasive insects spread and carry threats to swiss ecosystems. We use AI driven methods to measure, trap and decelerate them.
Contact person: Nicola Storni
BiologicalControl of the Japanese Beetle Using Automated “Attract and Infect” Traps
Project Overview
The Japanese beetle (Popillia japonica) is a newly arrived invasive pest in Europe that threatens crops, turf, and ornamental plants.
Native to Japan, it has spread rapidly in recent years and is now established
in parts of southern Switzerland and northern Italy. Its broad host range and
high reproductive rate make it one of the most concerning invasive insect
species currently affecting European agriculture and biodiversity.
This project explores biological control strategies for the Japanese beetle, focusing on environmentally friendly and sustainable methods that reduce the need for chemical insecticides. The initiative combines entomopathogenic fungi with automated field technologies to develop a new generation of intelligent, self-sustaining control traps.
The project is conducted in collaboration with Dr. Giselher Grabenweger at Agroscope and the IPM Popillia project[NK1] , within the framework of European efforts to manage invasive species through integrated pest management.
Objectives
Develop and test biological control methods using naturally occurring entomopathogenic fungi.Design an automated “attract and infect” trap that lures beetles, inoculates them with fungal spores, and releases them to spread the pathogen within the population.Create a low-cost, low-power mechanical system suitable for long-term outdoor use.Integrate LoRa-based remote monitoring to track trap activity, motion, and temperature to ensure optimal fungal viability.Address challenges related to fungal spore longevity, power efficiency, and scalability for large-area deployment.
Contribution ETH Zurich
The Crop Science Group at ETH Zurich is developing the prototype automated dispersal trap, focusing on its electronic, mechanical, and data-acquisition components.
The system uses motion detection and low-power communication technologies to monitor trap function and environmental conditions in real time. By maintaining spore viability and minimizing maintenance requirements, the design aims to make biological control feasible on a wide scale.
Automatic Camera Trap for Early Warning ofthe Asian Hornet (Vespa velutina)
Project Overview
The Asian hornet (Vespa velutina) is an invasive predator of honeybees and biodiversity across Europe. This project investigates automated camera-based bait stations to detect V. velutina in real time, using low-power edge- processing hardware and machine learning models. Ten stations deployed across four Swiss cantons collected over 11,000 operating hours of monitoring in 2025. Only 15 confirmed V. velutina detections were made, while thousands of Vespa crabro events were recorded.
Objectives
• Evaluate continuous camera-based monitoring as an early-warning system.
• Develop and refine onboard image-classification models for hornet detection.
• Assess bait attractiveness and selectivity.
• Improve event-detection logic converting raw image streams into visitation data.
• Generate datasets for long-term model improvements.
Technical Approach
Hardware consists of solar-powered, tripod-mounted camera units with cellular connectivity. Software includes real‑time image classification (~14 frames/sec), multi-class insect detection, and sliding-window event logic designed for high recall of V. velutina.
Key Findings
• Only 15 validated V. velutina visits across three stations.
• Classification challenges due to bees, ants, beetles, and dark-colored wasps.
• Temporal patterns showed queen activity from April and worker emergence in July.
Contribution ETH Zurich
Development of hardware, on-device ML models, remote-update systems, and data-acquisition pipelines, enabling large-scale automated monitoring.
Next Steps
• Increase station mobility.
• Test alternative bait formulations, including protein-based options.
• Retrain and enhance models using newly collected datasets and object detection techniques.
