A Self-Prompted YOLOv11–SAM 2 Pipeline for Automatic Plant Disease Detection and Segmentation

Résumé

Plant disease detection and segmentation are essential to maintaining healthy crops and improving agricultural productivity. Using Artificial Intelligence (AI) for this task enables farmers and researchers to identify diseases early and take preventive action. However, accurately segmenting diseased regions remains challenging, as most methods require a large number of labeled images or manual guidance to train the model. This limits their scalability and practical use in real-world agricultural settings. To address this issue, we propose the YOLO_SAM model for plant disease detection and segmentation, using self-prompted mask generation. Experiments were conducted on a self-generated dataset containing samples of different leaf diseases. The YOLO11 model was first used to detect infected regions, and its bounding boxes were automatically passed to the Segment Anything Model (SAM), which generated detailed segmentation masks without manual input. Experimental results showed that YOLOv11s achieved the best performance with a mean Average Precision (mAP) of 0.845, outperforming YOLOv11n (0.825), YOLOv11l (0.807), and YOLOv11m (0.764). Based on this superior performance, YOLOv11s was chosen as the prompt generator for SAM2, enabling more accurate and reliable segmentation of disease regions. This combined approach enabled the model to generate precise lesion masks without manual prompting, allowing clear boundary extraction even for small or irregular disease spots.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0