An RGB-D Image Dataset for Lychee Detection and Maturity Classification for Robotic Harvesting

The dataset contains color (RGB) images collected under different weather conditions and at different time periods, with a resolution of 1280×1024. The images cover various lychee varieties, such as Nuomi, Feizi Xiao, Heiye, and Huaizhi. The dataset includes three distinct ripening stages: immature, semi-ripe, and mature, comprising a total of 11,414 images. These include 878 original RGB images, 8,780 enhanced RGB images, and 1,756 depth images. The images are organized into five folders based on collection date, such as Image_YoloLabel_250605_Outdoor_sunny. Each folder includes descriptions of the weather and indoor/outdoor scenes. Each image and its labeled file is numbered, and includes the data augmentation method, number of categories, and similarity score (e.g., ID-0001_cc_unripe-0_semi-ripe-0_ripe-2_sim-0.0016.jpg). These images are labeled with 9,658 pairs of tags for lychee detection and ripeness classification during robotic harvesting. To improve the consistency of annotation, three researchers independently annotated the data, and then a fourth reviewer summarized and verified their results.