Tomato Leaf Damage Progression
Description
This dataset provides 8,596 images of tomato leaves focusing on the progression of three diseases: Fusarium spp., Leaf Miner, and Early Blight. It is designed to support the development and evaluation of deep learning models for object detection and severity stage classification under real field conditions. The images are categorized into three severity levels (Initial, Intermediate, and Advanced), enabling the study of disease evolution. The dataset includes original field captures, manually cropped leaf instances, and synthetic images generated via Poisson blending to improve early-stage representation. Additionally, selective median filtering was applied to the Early Blight class to enhance data quality. The dataset follows the YOLO annotation format, allowing direct use in object detection frameworks and facilitating reproducibility in agricultural computer vision research.
Files
Steps to reproduce
Data Collection: A set of 3,900 images is presented including 2,930 captures of Fusarium spp., and leaf miner, and 970 images of Early Blight obtained from the public Tomato Leaf Disease Dataset. Individualization: Raw images were manually cropped to isolate single leaves, increasing the set to 4,630 images. Synthetic Generation: 600 synthetic images representing early-stage symptoms were created using Poisson blending. Labeling: 14,084 instances were labeled using Roboflow, categorizing three diseases (Fusarium, Leaf Miner, Early Blight) into three severity levels (Initial, Intermediate, Advanced). Preprocessing: - A selective Median Filter (5x5) was applied specifically to the Early Blight images using OpenCV to mitigate salt-and-pepper noise. All images were converted to RGB color space and saved via Pillow with a quality factor of 95. Augmentation: Field robustness was increased via rotation, brightness adjustment, resulting in a final dataset of 8,596 images and 20,291 annotations.