A dataset of litchi pest and disease images

This is the authors' abstract. We don't add key points for this paper.

Information-based prevention and control technologies for litchi pests and diseases play an important role in improving fruit quality and increasing yield in litchi production. Deep learning-based object detection methods are one of the key components of intelligent pest and disease management. However, the performance of such detection models is predominantly determined by the quality of the dataset used for training. However, currently litchi pest and disease image datasets are often limited in sample size and exhibit inconsistent image quality, which severely constrains the development of detection models for litchi pests and diseases. This study constructed a high-quality litchi pest and disease image dataset through the following rigorous process. First, high-resolution original images of litchi pests and diseases were acquired in situ within litchi orchards using image acquisition equipment. Subsequently, these raw images underwent cross-verification by plant protection experts and meticulous annotation by trained labelers to ensure accurate classification of pests and disease and eliminate low-quality samples. Finally, a high-quality dataset of 4,393 images was established, comprising six categories: algal spot, anthracnose, damage symptoms caused by the litchi leaf gall midge, sooty mold, obverse felt disease, and reverse felt disease. Ultimately, four object detection models—Faster RCNN, YOLOv5, YOLOv8, and YOLOv12—were selected to validate the effectiveness of the dataset. The dataset provides both image samples and corresponding XML annotation files, supporting applications such as image classification, object detection, and pest and disease recognition. It holds significant importance for advancing intelligent and environmentally friendly technologies for litchi pests and disease management.