Combined Citrus maxima (Pomelo) and Phyllanthus emblica (Indian Gooseberry) Leaf Dataset An Expert-Validated Balanced Resource for Deep Learning-based Disease Detection in Bangladesh

This article presents a comprehensive, high-resolution image dataset capturing the phenotypic variations and pathological conditions of two indigenous South Asian fruit species: Citrus maxima (Pomelo) and Phyllanthus emblica (Indian Gooseberry). Systematically collected to advance agricultural artificial intelligence and plant pathology research, the repository encompasses a total of 4,502 images, structured into two distinct subsets: 'Raw Images' containing 868 original high-resolution captures, and 'Augmented Images' comprising 3,634 processed samples. Data acquisition was conducted in homestead agro-forestry ecosystems in Tangail, Bangladesh, between January 13, 2025, and October 20, 2025, utilizing a Realme 7i smartphone (64MP) under uncontrolled, "in-the-wild" natural lighting conditions. The dataset is meticulously categorized into nine classes, including four specific disease symptoms of Pomelo (Fungal Curling, Yellow Mosaic, Leaf Hole, and Chewing Damage), two healthy growth stages of Pomelo (Early Stage and Healthy), and three phenological stages of Phyllanthus emblica (Healthy, Partially Defoliated, and Early Stage), which serve as morphological negative controls for cross-species validation. To mitigate seasonal class imbalance, a strategic data augmentation pipeline was implemented using the Albumentations library—applying geometric transformations, noise injection, and photometric distortions. While the raw images are preserved at their original resolution to retain fine-grained details, the augmented dataset is standardized to 512×512 pixels to facilitate robust deep learning model training. This open-access resource offers a balanced, expert-verified benchmark for developing age-invariant disease detection systems and precision agriculture solutions in tropical climates.