Image Dataset of Tea Chrysanthemums in Complex Outdoor Scenes

Tea chrysanthemums can provide many components that are beneficial to human health. However, the harvesting process is time-consuming and labor-intensive. In the future, tea chrysanthemums harvesting can be done by machines, the first step towards automated harvesting is the detection of tea chrysanthemums, which are highly dependent on the quantity and quality of datasets. In a natural environment, a strain of chrysanthemum can present multiple flower heads in different stages and sizes, and there is a lack of sufficient dataset, making it difficult to detect tea chrysanthemums in complex outdoor scenes. Consequently, we present a dataset about six types of tea chrysanthemums (Bo-chrysanthemum, Hangbaiju, Jinsihuangju, Wuyuanhuangju, Gongju, and Chuju) images with 81276 images (1080×1920 pixels). An image acquisition method based on the Mi 10 phone to capture the images of tea chrysanthemums, which can meet the requirement of detection of tea chrysanthemums in complex outdoor scenes. Compared with the pictures collected in a controlled environment in the past. This dataset contains five difficult-to-identify situations caused by complex outdoor conditions: (1) direct light, (2) backlight, (3) shadow, (4) occlusion, and (5) overlap. Moreover, 3000 images of each type tea chrysanthemums are labeled manually and saved in XML format, and this dataset is available for training and validation of machine learning models to detect tea chrysanthemums in complex outdoor scenes. Besides, this paper also provides 453 original images (5760×3240 pixels) and videos (1080P and 60FPS) of tea chrysanthemums for other researchers to perform image processing according to their requirement.

茶菊（Tea chrysanthemums）可为人体提供诸多有益健康的成分，但其采收过程耗时耗力。未来茶菊采收可实现机械化作业，而自动化采收的首要步骤便是茶菊检测，该任务高度依赖数据集的规模与质量。在自然环境中，一株茶菊可呈现多个不同发育阶段与尺寸的花头，且现有数据集规模不足，导致复杂户外场景下的茶菊检测难度极大。为此，本研究构建了涵盖6类茶菊的数据集：亳菊（Bo-chrysanthemum）、杭白菊（Hangbaiju）、金丝皇菊（Jinsihuangju）、婺源黄菊（Wuyuanhuangju）、贡菊（Gongju）以及滁菊（Chuju），共包含81276张分辨率为1080×1920像素的图像。本研究采用基于小米10（Mi 10）手机的图像采集方案获取茶菊图像，可满足复杂户外场景下的茶菊检测需求，相较于过往受控环境下采集的图像，本数据集包含5类由复杂户外环境导致的难识别场景：（1）直射光照、（2）逆光、（3）阴影、（4）遮挡、（5）重叠。此外，每类茶菊的3000张图像均已完成人工标注，标注文件以XML格式存储，该数据集可用于训练与验证面向复杂户外场景茶菊检测的机器学习模型。同时，本文还提供了453张分辨率为5760×3240像素的原始茶菊图像，以及1080P分辨率、60FPS帧率的茶菊视频，可供其他研究人员根据自身需求开展图像处理相关研究。

搜集汇总

背景与挑战

背景概述

该数据集是一个专注于茶菊在复杂户外场景下的图像数据集，包含六种茶菊的81,276张图像，专门设计用于支持自动化收获中的检测任务。其特点在于采集自五种挑战性户外条件（如直射光、背光等），并提供了18,000张手动标注的图像和额外原始数据，适用于训练和验证机器学习模型。