High-Resolution Dataset of Common Bean Leaf Diseases in Bangladesh

This dataset consists of 3,180 high-resolution images of Phaseolus vulgaris (common bean) leaves, collected from the Rangpur Division in Bangladesh. The dataset is divided into five classes: Angular Leaf Spot (ALS), Anthracnose (ANT), Mosaic Virus (MV), Rust, and Healthy Leaves. The images were captured under natural daylight conditions using mobile phones, ensuring real-world applicability for field-based disease detection.

The dataset includes the following class distribution:

Angular Leaf Spot (ALS): 635 images

Anthracnose (ANT): 671 images

Mosaic Virus (MV): 629 images

Rust: 625 images

Healthy Leaves: 620 images

The dataset is well-balanced, which prevents bias during training, making it ideal for machine learning applications. The high-resolution images (average 4020 × 3093 pixels) offer detailed views of the leaf surface, essential for fine-grained feature extraction. The images were captured with an A4 background to minimize environmental disturbances and ensure consistent quality across the dataset.

Accompanying the images is comprehensive metadata, which includes details on image resolution, file size, sharpness, noise, brightness, RGB values, and texture features like contrast and homogeneity. This metadata is useful for further analysis of disease-specific visual patterns and enables researchers to investigate disease classification under standardized conditions.

This dataset is particularly suited for training and validating deep learning models such as Convolutional Neural Networks (CNNs), Vision Transformers, and YOLO-based architectures for disease classification and severity detection. It provides a valuable resource for AI-based plant pathology research, with potential applications in precision agriculture, aimed at reducing crop yield losses due to late disease detection. By enabling early disease identification, the dataset supports sustainable agricultural practices and enhances food security.

The dataset is publicly available through Mendeley Data and comes in a ZIP archive containing the raw JPG images and corresponding CSV metadata files. The dataset is a reusable resource for machine learning-based plant disease detection, and it can be further augmented with other datasets for cross-regional benchmarking in disease monitoring and early intervention.