6WEED（玉米田6类杂草数据集 ）

针对真实场景下的具有相似特征的杂草数据集短缺，在中国新疆维吾尔自治区昌吉回族自治州昌吉市华兴农场进行杂草图片采集。数据图片是使用移动手机vivoZ5采集的RGB(红-绿-蓝)彩色图像，宽高比设置为4:3，分辨率为4096×3027，拍摄前关闭曝光、聚焦等各种功能，保障图片数据的原始性不被设备自带算法破坏。本研究中的数据集为了确保图像的多样性和变化性，以提高模型的鲁棒性，在自然光照条件和不同天气条件（例如，晴天、多云和阴天）下，在杂草的不同生长阶段和不同的田间地点，分别在早上、中午和下午进行数据图片采集。最终确定数据集为六种杂草(反枝苋、灰黎、龙葵、骆驼刺、马齿苋、田旋花)，共1990张图片，其中的反枝苋、灰黎、龙葵幼苗具有相似的外观，骆驼刺为新疆特有的杂草，可以为以后的针对杂草相似性科学研究提供数据支持。Fig. 1展示了带注释的图像示例，其中每行图像仅突出显示一个杂草类。采集的杂草图片在形状、纹理、颜色、土壤背景和田间光照条件下呈现出类别间及类别内部的多样性，这种多样性有助于数据集更贴近真实场景，这对于构建对成像条件具有鲁棒性的模型十分重要。

Aiming at the shortage of weed datasets with similar morphological features in real-world scenarios, weed image collection was conducted at Huaxing Farm, Changji City, Changji Hui Autonomous Prefecture, Xinjiang Uygur Autonomous Region, China. The images were collected as RGB (Red-Green-Blue) color images using a vivo Z5 mobile phone, with an aspect ratio of 4:3 and a resolution of 4096×3027. All functions such as exposure and focus were disabled before shooting to ensure the originality of the image data was not compromised by the built-in algorithms of the device. To ensure the diversity and variability of the images and improve the robustness of subsequent models, the dataset collection was carried out under natural lighting conditions and different weather conditions (e.g., sunny, cloudy, and overcast), at different growth stages of weeds, in different field locations, and at different times of the day including morning, noon, and afternoon. The finalized dataset contains a total of 1990 images covering six weed species: Amaranthus retroflexus, Chenopodium album, Solanum nigrum, Alhagi sparsifolia, Portulaca oleracea, and Convolvulus arvensis. The seedlings of Amaranthus retroflexus, Chenopodium album and Solanum nigrum have similar appearances, while Alhagi sparsifolia is a weed endemic to Xinjiang. This dataset can provide data support for future scientific research on weed similarity. Figure 1 shows annotated image examples, where each row highlights only one weed species. The collected weed images exhibit considerable intra-class and inter-class diversity in terms of shape, texture, color, soil background, and field lighting conditions. This diversity helps the dataset better align with real-world scenarios, which is crucial for developing models robust to varying imaging conditions.

搜集汇总

背景与挑战

背景概述

该数据集是一个专注于玉米田杂草检测的图像数据集，包含六种杂草类别（反枝苋、灰黎、龙葵、骆驼刺、马齿苋、田旋花），共1990张高分辨率RGB图像和2352个标注边界框。其特点在于数据采集条件多样（涵盖不同天气、光照和生长阶段），增强了模型在真实场景中的鲁棒性，且包含外观相似的杂草类别和新疆特有杂草骆驼刺，适用于杂草识别和相似性研究。