PubMed · 2026-07-02
Researchers built a lightweight AI model called PD-ViCo that can identify five brinjal (eggplant) diseases from photos with 99% accuracy, using a new dataset of field-harvested images from Bangladesh. The model also explains its decisions visually, making it practical for farmers to use in real conditions.
PD-ViCo achieved 99.12% classification accuracy and 97.76% F1-score across five disease classes, outperforming both standard Vision Transformer and Swin Transformer baselines.
A new dataset of 1,823 field-harvested brinjal images was created from real agricultural conditions in Bangladesh, covering Phomopsis Blight, Fruit and Shoot Borer, Fruit Cracking, Wet Rot, and Healthy samples.
Grad-CAM visualizations confirmed the model focuses on disease-affected regions of the fruit, providing interpretable evidence that supports trust in real-world deployment.
