GuavaLeafVision: A Labeled Dataset of Guava Leaf Diseases for Image-Based Detection

This dataset supports research in plant pathology and computer vision by providing a labeled image collection of guava leaf diseases, captured under real-world agricultural conditions. The original dataset includes 1,050 high-resolution images representing seven categories: Caterpillars, Cutting Weevil, Die Back, Healthy, Mealybug Pests, Red Rust, and Yellow Spot. Each class contains 150 images, ensuring class balance. The images were captured using two high-end smartphones (Samsung Galaxy S21+ and iPhone 11) in two guava gardens in Ashulia, Savar, Dhaka, Bangladesh. The original images have a resolution of 3024x4032 pixels, later resized to 480x480 for initial processing. To enhance generalizability and model robustness, the dataset was augmented using the following techniques: • Image Resizing: All images resized to 224x224 pixels. • Normalization: Pixel values scaled to the [0, 1] range. • Data Augmentation: Applied rotation, shearing, and flipping to increase variability and simulate real-life conditions. This augmentation expanded the dataset to 8,400 images, increasing the sample size and representation of disease diversity. It is ideal for training and evaluating deep learning models for plant disease classification, particularly in resource-constrained and agro-environmental contexts. Number of Classes: 7 Original Images: 1,050 Augmented Images: 8,400 Total Images: 9,450 Image Format (Original): .jpg Image Format (Augmented): .png Dimensions: 480x480 (Original) Dimensions: 224x224 (Augmented)本数据集提供了真实农业场景下采集的番石榴叶病害标注图像集，可支撑植物病理学与计算机视觉领域的相关研究。原始数据集包含1050张高分辨率图像，涵盖7个类别：毛虫虫害、切象甲病害、回枯病、健康叶片、粉蚧虫害、红锈病以及黄斑病，每类均包含150张图像，确保类别分布均衡。 该数据集的原始图像由两台高端智能手机（三星Galaxy S21+与iPhone 11）在孟加拉国达卡萨瓦阿舒利亚的两座番石榴园中拍摄完成。原始图像分辨率为3024×4032像素，后续被调整为480×480像素以开展初步处理。 为提升模型的泛化能力与鲁棒性，本数据集采用以下技术进行数据增强： • 图像缩放：将所有图像统一调整至224×224像素分辨率。 • 归一化：将像素值缩放至[0, 1]区间。 • 数据增强：通过旋转、剪切与翻转操作增加样本多样性，模拟真实农业环境中的各类场景。 本次数据扩增将数据集规模扩充至8400张图像，提升了样本总量与病害多样性的覆盖度。本数据集非常适合用于训练与评估植物病害分类深度学习模型，尤其适用于资源受限的农业环境场景。 类别数量：7类 原始图像数量：1050张 增强后图像数量：8400张 总图像数量：9450张 原始图像格式：.jpg 增强后图像格式：.png 图像尺寸（原始）：480×480像素 图像尺寸（增强后）：224×224像素