Colorado Potato Beetle Dataset and Detection for Monitoring and Management in Potato Fields

Abstract

The Colorado potato beetle (Leptinotarsa decemlineata) remains a significant threat to potato crops worldwide, imposing substantial economic losses and challeng- ing sustainable agricultural practices. Manual pest control methods are labor- intensive, inefficient, and often insufficient to prevent widespread infestations. To address this challenge, we propose automated pest detection, have developed a labeled dataset (POBED), and studied several object detection models. Particu- larly YOLOv6 and CO-DETR, demonstrated promising performance in identifying Colorado potato beetle (CPB) stages, with AP IoU =.50 = 72.7% for YOLOv6 and AP IoU =.50 = 79.2% for CO-DETR. Despite challenges with background elements and labeling inconsistencies, this research highlights the potential of such models for generating detailed infestation maps and guiding targeted pest control strategies. Further refinement and exploration, including integration with autonomous removal mechanisms, offer exciting avenues for enhancing pest man- agement efficiency and sustainability in agriculture