Standardized litchi leaf disease image dataset for deep learning applications

Standardized litchi leaf disease image dataset is a curated image dataset of healthy and diseased litchi (Litchi chinensis) leaves designed for computer vision and deep learning research in plant pathology. The dataset contains 11 classes of leaf conditions captured using a high-resolution smartphone camera under natural lighting on a clean white background to reduce visual noise and enhance symptom visibility. Raw images were pre-processed through a standardized pipeline that performs background removal, leaf-centric cropping and aspect ratio preserving resizing to generate three aligned versions of each sample: 512 × 512 pixels with a black background, 512 × 512 pixels with a white background and 224 × 224 pixels with a white background tailored for transfer learning models.

In addition to the original pre-processed images, this repository also includes an augmented dataset where each 224 × 224 sample is expanded using geometric and photometric transformations such as rotation, scaling, flipping, elastic deformation, Gaussian blur, cutout, negative transform and brightness variations. This results in 1–12 images per original sample and provides increased variability for training robust models. All images are RGB and each sample has been reviewed and labeled by agricultural experts to ensure reliable ground truth. The dataset is intended for reuse in tasks such as disease classification, image based plant phenotyping, segmentation of symptomatic regions and severity analysis as well as for benchmarking new computer vision methods on litchi leaf disease data and for teaching in machine learning, deep learning and digital agriculture.