发电厂卫星图像数据集
收藏资源简介：
该数据集包含美国境内 4454 座发电厂的卫星图像。图像提供了两种分辨率：1 米（带有近红外波段的 4 波段美国国家农业影像计划（NAIP）图像）以及 30 米（陆地卫星 8 号图像，经全色锐化后分辨率达 15 米）。美国国家农业影像计划图像仅适用于美国地区，而陆地卫星 8 号图像全球皆可获取。该数据集对于计算机视觉工作、机器学习以及能源和环境分析可能颇具价值。 此外，还提供了每张图像中发电厂空间范围具体位置的标注信息。这些标注是通过众包平台 “亚马逊土耳其机器人”（Amazon Mechanical Turk）收集的，针对每张图像使用了多名标注人员以确保质量。图像数据来源、标注工具以及创建该数据集的团队的相关链接都包含在 “参考文献” 部分。 若要进一步了解这些数据，请查阅 “发电厂卫星图像数据集概述.pdf” 文件。若想下载数据示例而无需下载整个数据集，可下载 “sample.zip” 文件，其中包含两座发电厂示例以及它们各自对应的美国国家农业影像计划图像、陆地卫星 8 号图像和二进制标注信息。 注意：在标准图像查看软件中查看美国国家农业影像计划图像时，其可能看起来有些 “褪色”，这是因为除了标准的红、绿、蓝（RGB）数据外，它还包含了近红外波段。 资助情况 该数据集由杜克大学的学生和教职员工创建，获得了以下项目的资金支持：能源倡议项目、信息倡议的数据 + 计划、能源领域的巴斯连接项目、普拉特工程学院以及尼古拉斯环境政策解决方案研究所。
This dataset contains satellite imagery of 4,454 power plants located within the United States. The imagery is available in two resolutions: 1-meter (4-band U.S. National Agriculture Imagery Program (NAIP) imagery with a near-infrared band) and 30-meter (Landsat 8 imagery, which was pan-sharpened to a 15-meter resolution). NAIP imagery is only available for regions within the U.S., while Landsat 8 imagery is accessible globally. This dataset holds significant value for computer vision tasks, machine learning, and energy and environmental analytics. In addition, annotation information detailing the exact spatial extents of the power plants within each image is provided. These annotations were collected via the crowdsourcing platform Amazon Mechanical Turk, with multiple annotators assigned to each image to ensure quality. Relevant links to the image data sources, annotation tools, and the team that created this dataset are included in the "References" section. For further information about this dataset, please refer to the "Power Plant Satellite Imagery Dataset Overview.pdf" file. To download sample data without obtaining the entire dataset, you can download the "sample.zip" file, which contains examples of two power plants along with their corresponding NAIP imagery, Landsat 8 imagery, and binary annotation masks. Note: When viewed in standard image viewing software, NAIP imagery may appear somewhat "faded", as it contains a near-infrared band in addition to standard red, green, and blue (RGB) data. Funding This dataset was created by students and faculty at Duke University, with financial support from the following initiatives: the Energy Initiative, the Data + Initiative of the Information Initiative, the Bass Connections Project in Energy, the Pratt School of Engineering, and the Nicholas Institute for Environmental Policy Solutions.
