Abstract
Cydia pomonella L. (Lepidoptera: Tortricidae) is a major pest in pome fruit production, requiring accurate monitoring and timely interventions. This study provides key insights into trap design and performance validation and supports the integration of automated technologies into sustainable pest management tactics. We present the development and 3-year field validation of a fully automated prototype trap, equipped with a camera and a YOLOv8-based object detection model, for remote identification and counting of C. pomonella. In a laboratory setting, the YOLOv8-based system outperformed a previously published rule-based system (ImageJ1 + CNN). The manually evaluated proportion of correct detections and proportion of detections found (precision = 0.77 and recall = 0.83, respectively) indicated strong model performance. A moderate decline under field conditions was observed (precision = 0.65, recall = 0.63). Over six field experiments between 2022 and 2024, the prototype trap performed comparably to or better than commercial delta traps. No significant differences were observed among C. pomonella captures in white, orange, red, or green traps (with a mean cumulative capture per trap = 13.25), but white traps captured significantly more non-target insects. Trap entrance size also influenced capture performance under specific field conditions, and in one instance, the narrow-opening prototype captured significantly more C. pomonella than the wide-opening prototype (mean captures of 21.36 and 10.71, respectively). The system offers a reliable, effective, scalable, and automated solution for C. pomonella monitoring.