MSTAR（1996）
There have been many introductions on the Internet.
As a general library for SAR image automatic target recognition (SAR ATR) research, the MSTAR database is widely used by many scholars. The experimental data uses the actual SAR ground stationary target data released by the MSTAR program supported by the US Defense Advanced Research Projects Agency (DARPA). Whether it is domestic or international, the research on SAR image target recognition is basically based on this data set. Expanded. The sensor used to collect this data set is a high-resolution spotlight synthetic aperture radar with a resolution of 0.3m×0.3m . Working in the X-band , the polarization used is HH polarization . Pre-processing is performed on the collected data, and a slice image with a pixel size of 128×128 including various targets is extracted from it. Most of the data are SAR slice images of stationary vehicles, including target images acquired by various vehicle targets at various azimuth angles.
Source: Summary of MSTAR radar dataset
OpenSARShip（2017）
Download address: http://opensar.sjtu.edu.cn/
OpenSAR is an open SAR image management and processing platform developed by Advanced Sensing Technology Center (AST) of Shanghai Jiaotong University, which is used for SAR image reading, processing, visualization and algorithm testing. SAR image management and algorithm testing are the main tasks of OpenSAR.
OpenSAR supports importing various SAR data sources, such as TerraSAR-X, RADARSAT 1/2, COSMO-SkyMed, etc. Users can search and view SAR image data through this platform. OpenSAR supports registration of various algorithms such as image denoising, scene classification, object detection, object recognition, change detection, etc. Users can search, configure and execute these algorithms through the platform, and a complete test report will also be provided to users.
High-Resolution SAR Ship Detection Dataset (2019)
High Resolution SAR Ship Detection Dataset-1.0
Sun Xian, Wang Zhirui, Sun Yuanrui, et al. AIR-SARShip-1.0: High Resolution SAR Ship Detection Dataset[J]. Journal of Radar, to be published. doi: 10.12000/JR19097
Download link: https://radars.ac.cn/web/data/getData?newsColumnId=abd5c1b2-fe65-47f7-8ebf-990273a91a48
The high-resolution SAR ship detection data set-1.0 (AIR-SARShip-1.0) released the first batch of 31 images, the image resolution includes 1m and 3m, the imaging mode includes spotlight and strip, and the polarization method is monopole The scene types include ports, islands and reefs, and sea surfaces with different levels of sea conditions. The targets cover nearly a thousand ships of more than ten categories such as transport ships, oil tankers, and fishing boats.
The image size is about 3000×3000 pixels, and the image format is Tiff, single channel, and 8/16-bit image depth. The annotation file provides the length and width dimensions of the corresponding image, the category of the annotation target, and the position of the annotation rectangle.
High Resolution SAR Ship Detection Dataset-2.0
Sun Xian, Wang Zhirui, Sun Yuanrui, et al. AIR-SARShip-1.0: High Resolution SAR Ship Detection Dataset[J]. Journal of Radar, 2019, 8(6): 852–862. doi: 10.12000/JR19097
Download link: https://radars.ac.cn/web/data/getData?newsColumnId=1e6ecbcc-266d-432c-9c8a-0b9a922b5e85
The high-resolution SAR ship detection data set-2.0 (AIR-SARShip-2.0) releases 300 images, the image resolution includes 1m and 3m, the imaging mode includes spotlight and strip, and the polarization mode is single polarization. The polarization mode is VV, and the scene types include ports, islands and reefs, and sea surfaces with different levels of sea conditions. The targets cover more than ten types of thousands of ships such as transport ships, oil tankers, and fishing boats.
The image size is about 1000×1000 pixels, and the image format is Tiff, single channel, 8/16-bit image depth. The annotation file provides the length and width dimensions of the corresponding image, the category of the annotation target, and the position of the annotation rectangle.
SSDD/SSDD+（2020）
In the data set SSDD, there are a total of 1160 images and 2456 ships, with an average of 2.12 ships per image, and the data set will continue to expand. Compared with the PASCAL VOC dataset with 20 types of targets, SSDD has fewer pictures, but the category is only ships, so it is sufficient to train the detection model.
Compared with the SSDD data, the SSDD+ data set changes the vertical frame into a rotated frame, which can realize the direction estimation of the target while completing the detection task.
RSDD-SAR: SAR ship oblique frame detection dataset (2022)
Xu Cong'an, Su Hang, Li Jianwei, et al. RSDD-SAR: SAR Ship Slant Frame Detection Dataset[J]. Journal of Radar, to be published. doi: 10.12000/JR22007.
RSDD-SAR: SAR Ship Slant Frame Detection Dataset
[Latest Achievement] RSDD-SAR: SAR Ship Slant Frame Detection Dataset (Video)
Under the guidance of Academician He You, Professor Xiong Wei and Professor Liu Yu, the team of Associate Professor Xu Cong'an from Naval Aeronautical University constructed the SAR ship oblique frame detection data set RSDD- SAR, the data set consists of 84 scenes of Gaofen-3 data, 41 scenes of TerraSAR-X data slices and 2 scenes of uncut large images, a total of 127 scenes of data, including multiple imaging modes, multiple polarization methods, and multiple resolutions. There are 7000 slices and 10263 ship instances. In addition, through experiments and analysis of several commonly used optical remote sensing image slant frame detection algorithms and SAR ship slant frame detection algorithms, benchmark indicators are formed for reference by relevant scholars.
HRSID（2020）
Download address: https://github.com/chaozhong2010/HRSID
This dataset was released by Su Hao of the University of Electronic Science and Technology of China in January 2020. HRSID is a dataset for ship detection, semantic segmentation and instance segmentation tasks in high-resolution sar images. The dataset contains a total of 5604 high-resolution SAR images and 16951 ship instances. The ISSID dataset draws on the construction process of the Microsoft Common Objects in Context (COCO) dataset, including SAR images of different resolutions, polarization, sea state, sea area, and coastal ports. This dataset serves as a benchmark against which the researchers evaluate their methods. For HRSID, the resolutions of SAR images are: 0.5m, 1m, 3m.
SAR-Ship-Dataset Multi-source multi-scale SAR ship slice dataset
Reference: A SAR Dataset of Ship Detection for Deep Learning under Complex Backgrounds
Wang Y, Wang C, Zhang H, et al. A SAR dataset of ship detection for deep learning under complex backgrounds[J]. Remote Sensing, 2019, 11(7): 765. doi: 10.3390/rs11070765
Download address: https://radars.ac.cn/web/data/getData?dataType=SARGroundObjectsTypes
https://github.com/CAESAR-Radi/SAR-Ship-Dataset
This data set includes nearly 40,000 slices of SAR ship detection, and uses the domestic Gaofen-3 satellite and ESA Sentinel-1 satellite data. Image resolution covers 1.7m to 25m, polarization modes include HH, HV, VH and VV, imaging modes include ultra-fine strip mode, fine strip mode, fully polarized strip mode, strip scan mode and interference width Amplitude mode, detailed parameters are shown in Table 1. The data set scenarios include ports, near-shore, islands, and open seas, and types include various common ship targets such as oil tankers, bulk carriers, large container ships, and fishing boats.
The slice size is 256×256 pixels, and the format is three-channel grayscale image, 24-bit depth JPG. The annotation file is in TXT format, with one target marked in one line, respectively recording the ship type, normalized ship center position (column, row label), normalized ship width and normalized ship length, in line with Yolo series, PolarMask, Format requirements for mainstream detection networks such as SSD and Faster-RCNN.
FUSAR dataset
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.
Download address: https://radars.ac.cn/web/data/getData?dataType=FUSAR
The FUSARShip high-resolution ship dataset contains 15 main ship categories, 98 subcategories and many marine objects that are not ship objects. The data slices are taken from 126 original GF-3 remote sensing images, the polarization mode includes DH and DV, the resolution is 1.124m×1.728m, and the imaging mode is UFS mode, covering various seas, lands, coasts, rivers and islands Scenes.
This data set has accumulated 16144 slices, including 6252 ships matching AIS information, 2045 strong false alarms such as bright spots similar to ships, 1461 bridges and coastlines, 1010 coastal areas and islands, and 1967 complex sea clutter , 1,785 for common sea and 1,624 for land, suitable for ship detection and recognition in complex sea.
Large-Scale Multi-Class SAR Target Detection Dataset-1.0
Jie Chen, Zhixiang Huang, Runfan Xia, Bocai Wu, Lei Sheng, Long Sun, and Baidong Yao. Large-scale multi-class SAR image target detection dataset-1.0[OL]. Journal of Radars, 2022.
Download address: https://radars.ac.cn/web/data/getData?dataType=MSAR
The large-scale multi-class SAR target detection data set-1.0 (MSAR-1.0) includes a total of 28,449 detection slices, using the data of Haisi-1 satellite and Gaofen-3 satellite.
The polarization modes of MSAR-1.0 dataset include HH, HV, VH and VV. The data set scenarios include airports, ports, coastal areas, islands, open seas, urban areas, etc.; types include four types of targets: aircraft, oil tanks, bridges, and ships, consisting of 1,851 bridges, 39,858 ships, 12,319 oil tanks, and 6,368 aircraft composition.
SAR-ACD dataset
(As mentioned in the academic report of "Journal of Radar" some time ago, it is not clear whether it is open source)
SADD dataset
Download link: https://github.com/hust-rslab/SAR-aircraft-data