YOLO11-based deep learning and machine vision framework for automated detection and counting of Rhopalosiphum padi (Hemiptera: Aphididae).

The bird cherry-oat aphid, Rhopalosiphum padi (Linnaeus; Hemiptera: Aphididae), is a major cereal pest and a vector within the barley yellow dwarf virus complex. Manual scouting is labor-intensive and can be inconsistent, motivating rapid, image-based monitoring. We developed R. padi Count, an optimized deep learning framework built on Ultralytics YOLO11 for automated detection and image-level counting of R. padi in cluttered imagery. Starting from a YOLO11n baseline, we integrated an ADown downsampling module, Triplet Attention, and a Focusing Diffusion Pyramid Network module to improve small-object discrimination while limiting computational overhead. The model was trained and evaluated on smartphone images acquired under semi-controlled greenhouse conditions that captured variable backgrounds and occlusion. Relative to evaluated YOLO baselines, R. padi Count improved detection accuracy and reduced counting error, achieving a mean average precision at an IoU threshold of 0.50 (mAP50) of 92.97%, with an image-level mean absolute error of 1.86 aphids per image and a root mean squared error of 3.50 aphids per image. These results support the feasibility of practical, image-based aphid monitoring in the studied setting and provide a foundation for future evaluations of cross-scenario robustness and on-device performance in deployment contexts.