An Image Dataset for Analyzing Tea Picking Behavior in Tea Plantations

Tea is a significant economic product in our country, and tea plantation harvesting constitutes an essential agricultural activity. The tea plantation picking work is gradually moving towards intelligence and mechanization. As an active research field, artificial intelligence recognition technology is expected to identify the large-scale tea plantation picking work that is being promoted under the current situation, as well as the identification of tea plantation picking behavior. This series of work is inseparable from the construction of datasets, but there is still a large gap in the current tea plantation picking data. The different behaviors and tools of picking personnel in the tea plantation picking scene have not been well distinguished and identified. Based on this blank area, this work establishes a dataset from online video slices of tea plantation scenes. The dataset consists of 12,195 sliced picking images featuring five different types of behaviors( list as: 1) pick; 2) pick(machinery); 3) walk; 4) talk; 5) stand) under five distinct environmental conditions: 1) sunny; 2) overcast; 3) cloudy; 4) foggy; 5) rainy. All labels for the dataset are provided in COCO format for public use.

茶叶是我国重要的经济作物，茶园采摘作为核心农业生产活动，其作业正逐步向智能化、机械化方向发展。作为当前的活跃研究领域，人工智能识别技术有望适配当前规模化推广的茶园采摘作业场景，实现茶园采摘行为的精准识别。此类研究工作的开展离不开数据集的支撑，但当前茶园采摘相关数据集仍存在较大缺口：茶园采摘场景中，采摘人员的不同作业行为与使用工具尚未得到良好的区分与识别。针对这一研究空白，本研究从茶园场景的在线视频切片中构建了一款数据集。该数据集包含12195张切片采摘图像，涵盖五种不同的作业行为（依次为：1. 手工采摘；2. 机械采摘；3. 行走；4. 交谈；5. 站立），采集场景覆盖五种环境条件：1. 晴天；2. 阴天；3. 多云；4. 雾天；5. 雨天。该数据集的所有标注均采用COCO格式，可供公开使用。