Diagnóstico de severidad en fresa mediante segmentación dual con deep learning

Objective assessment of foliar disease severity is essential for efficient phytosanitary management in strawberry crops. In this study, an automated diagnostic system based on dual segmentation using the YOLOv12 architecture was developed to quantify the severity of leaf scorch. The approach employs two independent models: one that segments the entire leaf contour and another that delineates disease-affected areas. Both models were trained on manually annotated and artificially augmented datasets—460 images for the leaf model and 240 for the lesion model—and evaluated using precision, recall, and mAP@0.5 :0.95 metrics. The leaf model achieved an mAP@0.5 of 0.971, while the lesion model attained 0.588, demonstrating strong detection capability even for diffuse or incipient lesions. Disease severity was calculated as the ratio of lesion area to total leaf area, yielding continuous values ranging from 3.10% to 68.54%. Results show that this approach provides a robust, interpretable, and scalable tool for accurate disease monitoring, with potential applications in precision agriculture systems using drones or mobile devices. This method represents a significant step toward data-driven decision-making, contributing to the sustainability and profitability of strawberry cultivation.