Okra Leaf Bangladesh

The OkraLeafBD dataset contains 1,980 images of healthy and diseased okra leaves collected from four agricultural sites in Chandpur and Dinajpur districts of Bangladesh. The dataset comprises six classes: one healthy class (Fresh, 242 images) and five disease classes, including Insects Eaten Leaves (395 images), Leaf Curl (322 images), Leaf Spot (330 images), Predisposition (326 images), and Yellow Vein Mosaic Virus (365 images). All images were captured using Samsung A52 and Poco X5 smartphones under consistent lighting conditions with white backgrounds. All classes received base augmentation (flips, rotation, brightness/contrast adjustments, noise, blur). Three disease classes received additional specialized transformations: Leaf Curl (5× augmentation with elastic transforms for deformation), Insects Eaten (5× augmentation with coarse dropout for holes), and Yellow Vein Mosaic (4× augmentation with enhanced color for lighting variation). Other classes received 3× augmentation with base transformations only, prioritizing biologically meaningful variations. This strategy generated 7,739 augmented images from the original 1,980 images, prioritizing biologically meaningful variations over artificial class balancing.