FUSAR数据集

数据主编：徐丰 王海鹏 FUSARShip 高分辨率船只数据集，包含15个主要船舶类别、98 个子类别和许多非船舶目标的海洋目标。数据切片取自126幅原始高分三号遥感图像，极化模式包含DH和DV，分辨率为1.124m×1.728m，成像模式为 UFS 模式，覆盖了各种海、陆、海岸、河流和岛屿场景。 本数据集累16144个切片，其中包括与 AIS 信息匹配的船只 6252 张，类似船的亮点等强虚警 2045 张，桥及海岸线 1461 张， 沿岸区域及岛屿 1010 张，复杂海波杂波1967张，普通海面1785张，陆地1624张，适用于复杂海面的船只检测与识别工作。图1为FUSARShip船只切片样例。 图1. FUSARShip船只切片样例 数据集内图像的标注标准为以船舶目标最小外接离心圆的圆心为中点，向外扩充256个像素点，船舶切片大小固定，以512像素×512像素的切片形式存储。 图2 为FUSARShip数据集的分类框架。根节点为海洋对象，分为船舶和非船舶两个分支。船舶节点几乎包括所有种类的船舶，如货船、油轮、渔船等。其中一些节点具有子类节点，如货船节点由散货船、杂货船、集装箱船等组成。此外，非船节点有三个子节点，分别是陆地样本、海洋样本和强虚警样本(如浮标、风车、海上养殖场等)。其中，陆地节点包含典型的自然和人造近岸建筑，如桥梁和海岸线，沿海土地和岛屿等。 图2. FUSARShip数据集分类框架 该数据集由复旦大学电磁波信息科学教育部重点实验室实施、构建。作为一个开放的标准数据集，FUSARShip包含多种类型的海洋目标样本，为高分辨率、多维 SAR 海洋图像解译提供有力支持。可用于 SAR 船舶 ATR 算法的发展，评估和基准测试。 数据集使用说明见“FUSAR-Ship数据集使用说明”。 本期数据引用格式如下： Xiyue HOU,Wei AO,Qian SONG,Jian LAI,Haipeng WANG,Feng XU.FUSAR-Ship:building a high-resolution SAR-AIS matchup dataset of Gaofen-3 for ship detection and recognition[J].Science China(Information Sciences),2020,63(04):40-58.

Data chief editors: Feng XU, Haipeng WANG. The FUSARShip High-Resolution Vessel Dataset contains 15 major ship categories, 98 sub-categories, and numerous non-ship marine targets. The data slices are cropped from 126 original Gaofen-3 remote sensing images, with polarization modes including DH and DV, a resolution of 1.124m × 1.728m, and an imaging mode of UFS. It covers various sea, land, coastal, river, and island scenarios. In total, this dataset comprises 16,144 slices, including 6,252 ship images matched with AIS information, 2,045 strong false alarms such as ship-like bright spots, 1,461 images of bridges and coastlines, 1,010 images of coastal areas and islands, 1,967 images of complex sea wave clutter, 1,785 images of ordinary sea surfaces, and 1,624 images of land. This dataset is suitable for ship detection and recognition tasks in complex marine environments. Figure 1 shows examples of FUSARShip vessel slices. Figure 1. Examples of FUSARShip vessel slices. The annotation standard for images in this dataset is to take the center of the minimum enclosing circumcircle of a ship target as the midpoint, expand outward by 256 pixels, and the ship slices are stored in a fixed size of 512 × 512 pixels. Figure 2 shows the classification framework of the FUSARShip dataset. Figure 2. Classification Framework of the FUSARShip Dataset. The root node is "Marine Objects", which is divided into two branches: Ship and Non-Ship. The Ship branch covers nearly all types of vessels, including cargo ships, tankers, fishing boats, and more. Some nodes have subcategories; for instance, the cargo ship subbranch includes bulk carriers, general cargo ships, container ships, and other subtypes. Additionally, the Non-Ship branch has three sub-nodes: Land Samples, Marine Samples, and Strong False Alarm Samples (such as buoys, wind turbines, offshore aquaculture farms, etc.). The Land Samples subnode includes typical natural and artificial nearshore structures, such as bridges, coastlines, coastal lands, and islands. This dataset was developed and established by the Key Laboratory of Electromagnetic Wave Information Science, Ministry of Education, Fudan University. As an open standard dataset, FUSARShip contains a diverse set of marine target samples, providing robust support for high-resolution, multi-dimensional SAR marine image interpretation. It can be used for the development, evaluation, and benchmarking of SAR ship Automatic Target Recognition (ATR) algorithms. For dataset usage guidelines, please refer to "FUSAR-Ship Dataset Usage Instructions". The citation format for this dataset is as follows: Xiyue HOU, Wei AO, Qian SONG, Jian LAI, Haipeng WANG, Feng XU. FUSAR-Ship: building a high-resolution SAR-AIS matchup dataset of Gaofen-3 for ship detection and recognition[J]. Science China(Information Sciences), 2020, 63(04): 40-58.