YOLO-Based Deep Learning Pipeline for Detecting Nutrient Deficiency Symptoms in Roselle (Hibiscus sabdariffa L.)

This study presents a YOLO-based deep learning pipeline for detecting visual symptoms associated with calcium (Ca), potassium (K), iron (Fe), and phosphorus (P) deficiencies in Hibiscus sabdariffa L., together with healthy samples. A multi-device dataset collected over eight consecutive weeks under variable greenhouse conditions was used to evaluate YOLOv11, YOLOv12, and YOLOv13 architectures using Precision, Recall, F1-score, mAP@50, mAP@75, and mAP@[50–95]. YOLOv13 achieved the highest overall localization accuracy, particularly at mAP@75, while YOLOv12 provided the best Precision–Recall balance and YOLOv11 the most consistent weekly performance. These results demonstrate the feasibility of detecting nutrient-deficiency symptoms across different growth stages and acquisition conditions. The proposed workflow provides a reproducible, scalable, and cost-effective basis for developing low-cost nutritional monitoring tools using consumer-grade imaging devices, supporting precision agriculture and timely nutrient management.