canola_detection_dataset.zip

RGB images were obtained from three distinct datasets featuring canola in its early growth stage intermingled with weeds, including <b>T1 Miling</b> (T1_miling), <b>T2 Miling</b> (T2_Miling), and <b>York canola</b> (YC). These datasets were collected in Western Australia from distances ranging between 0.5 meters and 1.5 meters, captured at various angles. The images are processed as 500 by 500 pixel frames, comprising the core images along with their corresponding segmentation masks. Manual delineation of bounding boxes using the polygon tool was performed using the Make Sense annotation tool by Skalski, P. [GitHub: https://github.com/SkalskiP/make-sense/].For T1 and T2 Millet datasets, the images showcase canola plants in conjunction with rye grass, whereas the York canola dataset features images of canola alongside regrowth of blue lupin. Image collection involved the use of a smartphone for T1 and T2 and a Canon 600D DSLR camera for YC. If you are interested in the scripts for image processing and deep learning, they can be accessed at [GitHub: https://github.com/mikemcka/Canola_detection_dl/tree/main].In this repository you will find the RGB images and weed-crop-soil segmentation masks obtained using CIVE vegetation index.

本数据集的RGB图像源自3个针对早期生长阶段油菜与杂草混生场景的独立数据集，分别为<T1 Miling（T1_miling）>、<T2 Miling（T2_Miling）>及<York canola（YC）>。所有数据集均采集于西澳大利亚州，拍摄距离介于0.5米至1.5米之间，且采用多种拍摄角度。图像均被统一处理为500×500像素的图像帧，包含原始RGB图像及其对应的分割掩码（segmentation mask）。标注工作由Skalski P.借助其开发的Make Sense标注工具（GitHub地址：https://github.com/SkalskiP/make-sense/）完成，通过多边形工具手动绘制边界框。其中T1与T2 Miling数据集的图像呈现了油菜与黑麦草混生的场景，而York canola数据集则展示了油菜与再生蓝羽扇豆共生的画面。T1与T2数据集采用智能手机采集图像，YC数据集则使用佳能（Canon）600D数码单反相机拍摄。若需获取图像处理与深度学习相关脚本，可访问以下GitHub仓库：https://github.com/mikemcka/Canola_detection_dl/tree/main。该仓库中包含采用CIVE植被指数生成的RGB图像及杂草-作物-土壤分割掩码。