BanglaRiceLeaf: A Benchmark Dataset for Automated Rice Leaf Disease Detection and Health Classification in Bangladesh

Accurate and timely detection of rice leaf diseases is essential for maintaining crop health and improving yield, yet large-scale field monitoring remains challenging due to rapid disease progression and varying environmental and lighting conditions. Deep learning based automated detection systems offer an effective solution, but their performance depends heavily on the availability of diverse and well annotated datasets, while existing rice leaf datasets often lack sufficient variation in disease stages, rice varieties, image quality, and real world conditions. To address these limitations, we introduce BanglaRiceLeaf, a comprehensive dataset containing 4,152 high resolution images of rice leaves collected from July 2023 to July 2024 in the experimental fields of the Bangladesh Rice Research Institute (BRRI), Gazipur. Images were captured using the rear cameras of seven smartphones, including five iPhone 11 Pro Max and two iPhone 11 devices, under diverse lighting and weather conditions such as low and high light. All images are stored in JPG, JPEG, and PNG formats with a resolution of 1440 × 2560 pixels. The dataset is organized into five classes, representing four rice leaf diseases, Bacterial Leaf Blight, Bacterial Leaf Streak, Sheath Blight, and Leaf Blast, along with Healthy Leaf samples, and is carefully annotated and split into training, validation, and test sets, providing a reliable resource for developing and evaluating deep learning models for precision agriculture.

精准且及时地检测水稻叶片病害，是维持作物健康、提升单位产量的关键环节，但由于病害发展迅猛、田间环境与光照条件复杂多变，大规模田间监测仍存在诸多挑战。基于深度学习的自动化检测系统为该问题提供了可行的解决方案，但其性能高度依赖于多样化且标注完备的数据集。现有水稻叶片数据集往往在病害阶段、水稻品种、图像质量以及真实田间场景等维度的变异性不足。为破解上述局限，我们构建了BanglaRiceLeaf数据集：这一综合性数据集共包含4152张水稻叶片高分辨率图像，采集时间为2023年7月至2024年7月，采集地点位于孟加拉国水稻研究所（Bangladesh Rice Research Institute, BRRI）加济布尔试验田。图像由7款智能手机的后置摄像头拍摄，其中包括5台iPhone 11 Pro Max与2台iPhone 11设备，拍摄场景涵盖弱光、强光等多样光照与天气条件。所有图像均以JPG、JPEG以及PNG格式存储，分辨率统一为1440×2560像素。该数据集共划分为5个类别，涵盖4种水稻叶片病害：细菌性叶枯病（Bacterial Leaf Blight）、细菌性条斑病（Bacterial Leaf Streak）、纹枯病（Sheath Blight）以及稻瘟病（Leaf Blast），同时包含健康叶片样本；数据集已完成严谨标注，并划分为训练集、验证集与测试集，可为开发与评估用于精准农业的深度学习模型提供可靠的研究资源。