超棒的卫星遥感图片比赛
下面是利用机器学习和深度学习来对卫星图像和遥感图像做分类、分割、目标识别提取等任务的比赛列表,按照时间截止日期来排序。
- xView 2018 Detection Challenge (DIUx, Jul 2018)
目标检测 (60 类), 100万个实例, 0.3m 分辨率 Worldview-3 图像, COCO图像 数据格式,提前训练好的 Tensorflow 和 Pytorch 基准模型。 - CrowdAI Mapping Challenge (Humanity & Inclusion NGO, May 2018)
语义分割/实例分割(建筑物), RGB 图像, COCO 数据格式 - Open AI Challenge: Aerial Imagery of South Pacific Islands (Worldbank, May 2018)
物体检测 (4种树的种类), 语义分割 (2 中道路类别), 0.4m/0.8m 航拍图像, multiple AOIs in Tonga - DEEPGLOBE – 2018 Satellite Challange (CVPR, Apr 2018)
一共有道路提取,建筑物检测, 植被分类三个挑战。 - IEEE Data Fusion Contest 2018 (IEEE, -Mar 2018)
植被分类 (20 类) ,数据集有三种: 多谱线 LiDAR, 高光谱(1m), RGB 图像(0.05m) - Spacenet challenge – Round 3 (CosmiQ Works, Radiant Solutions, NVIDIA, Feb 2018)
道路提取, multiple city aois, 3/8 波段Worldview-3 imagery (0.3m), SpaceNet Challenge Asset Library - Statoil/C-CORE Iceberg Classifier Challenge (Statoil/C-CORE, Jan 2018)
图像识别 (预测视野图片是否有船或者冰山), 2-波段HH/HV polarization SAR imagery, Kaggle kernels - Functional Map of the World Challenge (IARPA, Dec 2017)
物体检测 (63 类), 100万个实例, 4/8 波段, COCO 数据格式, baseline algorithms - Urban 3D Challenge (USSOCOM, Dec 2017) 建筑物占地面积检测, 50cm 2D RGB 正的 photos and 3D data generated from satellite imagery, 3 cities, open source software for winning solutions, data hosted on SpaceNet Challenge Asset Library
- NIST DSE Plant Identification with NEON Remote Sensing Data (inria.fr, Oct 2017)
提取树的位置、种类和冠状参数, 高光谱(1m), RGB 图像 (0.25m), 激光雷达点云和冠层高度模型。 - Planet: Understanding the Amazon from Space (Planet, Jul 2017)
图像识别(Predict 1 of 13 land cover and 1 of 4 cloud condition labels per image chip), Amazonian rainforest, 4 band (RGB-NIR, 3-5m), Kaggle kernels - Spacenet challenge – Round 2 (CosmiQ Works, Radiant Solutions, NVIDIA, May 2017)
建筑物提取, multiple city aois, 3/8band Worldview-3 imagery (0.3m), SpaceNet Challenge Asset Library - DSTL Satellite Imagery Feature Detection challenge (Dstl, Feb 2017)
目标检测 & 分类(10 类). 3 波段(RGB) and 16 波段 Worldview 3 imagery (0.3m – 7.5m), Kaggle kernels. - Spacenet challenge – Round 1 (CosmiQ Works, Radiant Solutions, NVIDIA, Jan 2017)
建筑物提取, Rio de Janeiro, 3/8 波段 Worldview-3 imagery (0.5m mosaic), SpaceNet Challenge Asset Library - Multi-View Stereo 3D Mapping Challenge (IARPA, Nov 2016)
Development of a Multi-View Stereo (MVS) 3D mapping algorithm that can convert high-resolution satellite images to 3D point clouds. - Draper Satellite Image Chronology (Draper, Jun 2016)
Predict the chronological order of images taken at the same locations over 5 days, Kaggle kernels - Inria Aerial Image Labeling (inria.fr)
语义分割 (建筑物), 多城市研究区域, 航拍图像 (0.3m)
相关术语:
UAV : unmanned aerial vehicle 无人驾驶航空器
AoI :(area of interest),感兴趣区。 在遥感与地理信息系统的一些软件中对研究区域的叫法。
Worldview-3 image:Worldview-3是一个卫星的名称。Worldview-3 image就是这个卫星拍的图像。
比赛收集来自:https://github.com/christophrieke/awesome-satellite-imagery-competitions