Research on a machine vision-based electro-killing pheromone-baited intelligent agricultural pest monitoring method.

Abstract

The annual global economic losses from pest-induced crop damage are substantial and difficult to quantify. Real-time monitoring of pest dynamics and timely control strategies are crucial for food security. Among the primary monitoring techniques, sex pheromone-baited trapping technology is instrumental in the detection and management of agricultural pests. To address existing limitations-such as manual insect collection and counting in conventional traps, inaccuracies in photoelectric counting devices, and the requirement for manual replacement of sticky boards in image-based traps-an advanced agricultural pest monitoring system utilizing sex pheromone bait was designed and developed in this paper. The system integrates smart electro-killing pheromone traps, a pest detection model, and a pheromone monitoring platform. Male pests attracted to pheromones are neutralized by an electric grid and deposited on an image acquisition platform. A network camera captures images of the pests, which are processed by a YOLOv9-TrapPest detection model to identify and quantify them. This model incorporates an AKConv module to enhance feature extraction, reducing false detections from limb separation. The CBAM-PANet structure improves detection rates of sticky pests, while the FocalNet module optimizes fine-grained feature capture, excluding non-target pests. The YOLOv9-TrapPest model outperforms other detection models, achieving 97.5% average precision and 98.3% mAP50 for detecting seven pest species. Furthermore, a pest pheromone monitoring platform displays the images and identification results, supporting pest control decisions. This system incorporates automated functions for pest trapping, killing, counting, and clearing, thereby achieving complete automation in the monitoring of pests attracted by sex pheromones.