Part I—Modernizing Nemabot: AI-Supported Identification of Frankliniella occidentalis damage for enhanced biological control efficiency

Abstract

Western Flower Thrips (Frankliniella occidentalis) is a significant agricultural pest causing substantial economic losses by damaging crops and acting as a vector for plant diseases. Traditional pest control methods relying on chemical pesticides pose environmental and health risks, necessitating alternative solutions. Entomopathogenic nematodes (EPNs) have emerged as a promising biological control agent. This study presents an AI-supported precision application system, Nemabot, designed to optimize EPN deployment based on thrips-induced bean leaf damage. In this study, agricultural disease detection was performed using the Multi-Otsu Thresholding method integrated into deep learning-based object detection and segmentation algorithms. The developed method enhances segmentation accuracy through image processing techniques, thereby increasing the precision in identifying infested regions. The model used in the study was optimized with a YOLO-based architecture during training and reinforced with various data augmentation techniques for segmenting bean leaves. The model's performance evaluation yielded mAP0.5 values of B: 0.9481 and M: 0.94981, while mAP0.5:0.95 values were B: 0.90887 and M: 0.90887. The precision and recall values were 1.0 and 0.99975, respectively, indicating the model's high sensitivity. Additionally, the low values of box_loss, segmentation_loss, and objectness_loss demonstrate that the model maintains a minimal error rate. The proposed approach offers higher accuracy and sensitivity than conventional segmentation methods, contributing significantly to agricultural disease detection applications.