TomatoDiverse：An Open Dataset for Industrial Tomato Detection in Complex Natural Environments

Industrial tomatoes are one of the important agricultural products worldwide and play a key role in the agricultural economy. To promote the application of computer vision in smart agriculture, due to the lack of suitable datasets, we focus on building a dataset focused on industrial tomato target detection in complex natural environments. The data is collected from the industrial tomato planting base in Hejing County, Bayingolin Mongolian Autonomous Prefecture, Xinjiang Uygur Autonomous Region. It covers 1502 industrial tomato images with a resolution of 4000x3000 pixels captured by cameras in complex natural environments, taking into account changes in illumination 、distance、occlusions and other challenges in various natural environments. The images are all in jpg format and annotated using LabelImg data annotation software. Using Pascal VOC XML format and YOLO format, we provide accurate tomato bounding box information for each image. The database is divided into three levels. The total folder contains three sub-folders, named smooth illumination, counter illumination and back illumination respectively. These three folders represent the first level, each of which contains two subfolders (representing the second level), named sparse and dense. Each folder in the second level contains four sub-folders (representing the third level), named cover, cover_labels, no_cover, no_cover_labels. The pictures contained in each folder have the same name labels matching them. This data resource was established to support the research on industrial tomato target detection algorithms and promote the development of smart agriculture. This dataset can be used to delve into research on intelligent control systems, automated harvesting robots, maturity assessment, and yield estimation.