高分辨率遥感图像细粒度目标识别数据集（FAIR1M_V1.0）

该数据集含15000余幅分辨率优于1米、尺寸从上千到上万像素不等的图像，具有100多万精细化标注、多角度分布的目标，场景覆盖全球上百个典型城市、乡镇，以及常用机场、港口等，其中来自我国自主产权高分系列卫星的数据占比超过80%，标注结果均经权威判读专家确认。
　　根据遥感应用的实际需求，该数据集将目标（包含飞机、船舶、车辆、球场和道路等）进行类别的细分，包括37个细粒度类别。例如，依据型号将飞机细分为波音式（波音737、747、777和787等）、空客式（空客220、321、330和350等）以及国产式飞机（C919和ARJ21等）；将船舶细分为液货船、干货船、渔船、邮轮、拖船和工程船等；对于车辆、球场和道路等要素也分别按照功能、尺寸等细分至多个类型。
　　同时，考虑到不同时间拍摄的遥感图像中蕴含更为丰富的目标关联信息，该数据集还提供了针对同一区域、不同时相的数据，是一套多时相、多分辨、多要素的遥感图像标准化样本集。

This dataset contains more than 15,000 images with a spatial resolution better than 1 meter and pixel sizes ranging from thousands to tens of thousands. It includes over 1 million finely annotated, multi-angle distributed objects, covering hundreds of typical cities, towns, as well as common airports, ports and other scenarios worldwide. Data from China's independently developed Gaofen series satellites accounts for more than 80% of the total dataset, and all annotation results have been verified and confirmed by authoritative remote sensing interpretation experts.

Based on the practical requirements of remote sensing applications, this dataset classifies the target objects (including aircraft, ships, vehicles, stadiums, roads and others) into 37 fine-grained categories. For example, aircraft are subdivided into Boeing-type (Boeing 737, 747, 777, 787, etc.), Airbus-type (Airbus 220, 321, 330, 350, etc.) and domestic-made aircraft (C919, ARJ21, etc.) based on their models; ships are subdivided into liquid cargo ships, dry cargo ships, fishing boats, cruise ships, tugboats and engineering ships, etc.; elements such as vehicles, stadiums and roads are also subdivided into multiple types according to their functions, sizes and other relevant criteria.

Meanwhile, considering that remote sensing images captured at different times contain richer target correlation information, this dataset also provides data of the same region at different temporal phases, making it a standardized remote sensing image dataset with multi-temporal, multi-resolution and multi-element characteristics.

搜集汇总

背景与挑战

背景概述

高分辨率遥感图像细粒度目标识别数据集（FAIR1M_V1.0）是一个包含15000余幅高分辨率遥感图像的大规模数据集，具有100多万精细化标注目标，覆盖37个细粒度类别。该数据集具有多时相、多分辨、多要素的特点，场景覆盖全球上百个典型区域，特别适合用于遥感图像细粒度目标识别研究。