Amrapali Mango Fruit Diseases: A Cultivar-Specific Image Dataset for Computer Vision and Deep CNN Classification

This dataset comprises a carefully curated set of labeled images for the Amrapali variety of mango fruits, specifically for the cultivar-specific detection of diseases using computer vision and convolutional neural networks (CNNs). The primary aim of this dataset is to aid research related to the automatic classification of diseases with zero inter-cultivar variance; hence, the study is limited to the Amrapali variety, which is commonly found in South Asia.
The dataset comprises a cumulative total of 3,500 high-quality images of fruits belonging to seven clearly distinguishable classes: Healthy, Anthracnose, Bacterial Canker, Scab, Powdery Mildew, Sooty Mould, and Stem End Rot, with each category having a fixed set of 500 images for fair training and testing of deep learning algorithms. The images were captured using consumer-grade mobile camera devices simulating practical field setups and natural lighting intensities in actual orchards and post-harvest setups.
The disease categories included in the study and development stages were determined based on the prevalence rate, distinctness, and importance with regard to fruit disease classification. Disorders, whether physiological or disease-related, that lack distinctness or can easily pass unnoticed from visual perception, especially in disease categorization concerning fruit, also did not take part in the study. The smaller categories undergo augmentation, including safe transformations such as rotation, flipping, and adjustment of brightness, while larger categories undergo downsampling.
All images are stored in class-wise folders and are systematically renamed based on a common naming convention for easy reproducibility and use. The dataset is meant for use in disease classification, transfer learning, comparison of deep CNN models for classification, and other comparative studies of precision agriculture models. By focusing on a single cultivar, this dataset enables the development of robust cultivar-specific disease detection models and contributes to reproducible research in agricultural computer vision.