D3096: Can computers effectively detect crop pests? Evaluating a deep-learning model for watermelon pest detection on sticky cards

Traditional pest monitoring is labor-intensive and prone to human error, underscoring the need for automated solutions. This study evaluated the performance of two deep-learning object detection models, YOLOv11n and RT-DETR in detecting striped and spotted cucumber beetles on yellow sticky cards used in watermelon fields. Sticky cards were collected across two growing seasons at the University of Delaware farms and annotated for four insect categories: cucumber beetles, lady beetles, and pigweed flea beetles. Images were preprocessed into 512×512 patches and augmented to balance underrepresented classes. Both models were trained using optimized hyperparameters and tested on unseen data. Results showed high detection accuracy (96–100%), with both models successfully distinguishing cucumber beetles from visually similar insects and showing minimal counting errors (MAE ≈ 0.05; RMSE ≈ 0.12). These findings demonstrate the feasibility of integrating computer vision into automated Integrated Pest Management (IPM) programs, paving the way for real-time field deployment and drone-assisted monitoring in precision agriculture.