A Multi-Fruit Leaf Image Dataset for Disease and Growth Stage Classification in Real-World Agricultural Environments

Accurate and scalable plant disease recognition remains a critical challenge in precision agriculture, particularly in resource-constrained South Asian environments. To address this gap, a comprehensive and heterogeneous multi-fruit leaf dataset is introduced to support automated plant pathology and growth stage monitoring. The dataset encompasses four economically vital species: Guava, Jackfruit, Mango, and Olive, captured across diverse field environments in Bangladesh (Rajbari, Savar, and Daffodil Smart City). Utilizing five distinct smartphone sensors to ensure hardware-agnostic robustness, a total of 2,308 original high-resolution images were collected. To address class imbalance and facilitate deep learning requirements, the dataset was expanded via geometric and photometric augmentation to 18,000 images, standardized to a 512×512 resolution. A key differentiator of this work is the inclusion of 18 granular classes, categorizing not only specific diseases (e.g., Anthracnose, Blight) but also critical physiological stages such as Young and Senescent. This dataset serves as a foundational resource for developing lightweight, edge-AI-driven mobile applications for real-time, in-the-wild agricultural diagnostics.