logo

SCIENCE CHINA Information Sciences, Volume 63 , Issue 4 : 140303(2020) https://doi.org/10.1007/s11432-019-2772-5

FUSAR-Ship: building a high-resolution SAR-AIS matchup dataset of Gaofen-3 for ship detection and recognition

More info
  • ReceivedOct 7, 2019
  • AcceptedJan 1, 2020
  • PublishedMar 10, 2020

Abstract

Gaofen-3 (GF-3) is China's first civil C-band fully polarimetric spaceborne synthetic aperture radar (SAR) primarily missioned for ocean remote sensing and marine monitoring. This paper proposes an automatic sea segmentation, ship detection, and SAR-AIS matchup procedure and an extensible marine target taxonomy of 15 primary ship categories, 98 sub-categories, and many non-ship targets. The FUSAR-Ship high-resolution GF-3 SAR dataset is constructed by running the procedure on a total of 126 GF-3 scenes covering a large variety of sea, land, coast, river and island scenarios. It includes more than 5000 ship chips with AIS messages as well as samples of strong scatterer, bridge, coastal land, islands, sea and land clutter. FUSAR-Ship is intended as an open benchmark dataset for ship and marine target detection and recognition. A preliminary 8-type ship classification experiment based on convolutional neural networks demonstrated that an average of 79% test accuracy can be achieved.


Acknowledgment

This work was supported in part by National Key RD Program of China (Grant No. 2017YFB0502- 703) and National Natural Science Foundation of China (Grant Nos. 61991422, 61822107).


Supplement

Tables S1 and S2.


References

[1] Xu F, Jin Y Q, Moreira A. A Preliminary Study on SAR Advanced Information Retrieval and Scene Reconstruction. IEEE Geosci Remote Sens Lett, 2016, 13: 1443-1447 CrossRef Google Scholar

[2] Paes R L, Lorenzzetti J A, Gherardi D F M. Ship Detection Using TerraSAR-X Images in the Campos Basin (Brazil). IEEE Geosci Remote Sens Lett, 2010, 7: 545-548 CrossRef Google Scholar

[3] Liu Y, Yao L, Xiong W, et al. Fusion detection of ship targets in low resolution multi-spectral images. In: Proceedings of IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Beijing, 2016. 6545--6548. Google Scholar

[4] Shuai T, Sun K, Wu X, Zhang X, et al. A ship target automatic detection method for high-resolution remote sensing. In: Proceedings of IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Beijing, 2016. 1258--1261. Google Scholar

[5] Wang W, Fu Y, Dong F. Semantic segmentation of remote sensing ship image via a convolutional neural networks model. IET Image Processing, 2019, 55: 1016-1022 CrossRef Google Scholar

[6] Xu F, Wang H, Song Q, et al. Intelligent ship recongnition from synthetic aperture radar images. In: Proceedings of IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium, Valencia, 2018. 4387--4390. Google Scholar

[7] An Q, Pan Z, You H. Ship Detection in Gaofen-3 SAR Images Based on Sea Clutter Distribution Analysis and Deep Convolutional Neural Network.. Sensors, 2018, 18: 334 CrossRef PubMed Google Scholar

[8] Ma M, Chen J, Liu W. Ship Classification and Detection Based on CNN Using GF-3 SAR Images. Remote Sens, 2018, 10: 2043 CrossRef ADS Google Scholar

[9] Pan Z, Liu L, Qiu X. Fast Vessel Detection in Gaofen-3 SAR Images with Ultrafine Strip-Map Mode.. Sensors, 2017, 17: 1578 CrossRef PubMed Google Scholar

[10] Wang Y, Wang C, Zhang H. Automatic Ship Detection Based on RetinaNet Using Multi-Resolution Gaofen-3 Imagery. Remote Sens, 2019, 11: 531 CrossRef ADS Google Scholar

[11] Guo Q, Wang H, Kang L, et al. Aircraft target detection from spaceborne SAR image. In: Proceedings of IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2019. 1168--1171. Google Scholar

[12] Fu S, Xu F, Jin Y-Q. Reciprocal translation between SAR and optical remote sensing images with cascaded-residual adversarial networks. 2019,. arXiv Google Scholar

[13] Shao Z, He J, Feng S. Extraction of a target in sea clutter via signal decomposition. Sci China Inf Sci, 2020, 63: 129301 CrossRef Google Scholar

[14] Long T, Liang Z, Liu Q. Advanced technology of high-resolution radar: target detection, tracking, imaging, and recognition. Sci China Inf Sci, 2019, 62: 40301 CrossRef Google Scholar

[15] Ao W, Xu F, Li Y. Detection and Discrimination of Ship Targets in Complex Background From Spaceborne ALOS-2 SAR Images. IEEE J Sel Top Appl Earth Observations Remote Sens, 2018, 11: 536-550 CrossRef ADS Google Scholar

[16] Bandiera F, Orlando D, Ricci G. Advanced radar detection schemes under mismatched signal models. In: Synthesis Lectures on Signal Processing, , Morgan & Claypool, 2009. 1--105. Google Scholar

[17] Ge Liu , Yasen Zhang , Xinwei Zheng . A New Method on Inshore Ship Detection in High-Resolution Satellite Images Using Shape and Context Information. IEEE Geosci Remote Sens Lett, 2014, 11: 617-621 CrossRef ADS Google Scholar

[18] Chen H, Wang Q, Shen Y. Decision tree support vector machine based on genetic algorithm for multi-class classification. J Syst Eng Electron, 2011, 22: 322-326 CrossRef Google Scholar

[19] Deng J, Dong W, Socher R, et al. ImageNet: a large-scale hierarchical image database. In: Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Miami, 2009. 248--255. Google Scholar

[20] Everingham M, Van Gool L, Williams C K I. The Pascal Visual Object Classes (VOC) Challenge. Int J Comput Vis, 2010, 88: 303-338 CrossRef Google Scholar

[21] Liu N, Cao Z, Cui Z. Multi-Scale Proposal Generation for Ship Detection in SAR Images. Remote Sens, 2019, 11: 526 CrossRef Google Scholar

[22] Xia G, Bai X, Ding J, et al. DOTA: a large-scale dataset for object detection in aerial images. Computer science: computer vision and pattern recognition. 2017,. arXiv Google Scholar

[23] Zhu H, Chen X, Dai W, et al. Orientation robust object detection in aerial images using deep convolutional neural network. In: Proceedings of 2015 IEEE International Conference on Image Processing (ICIP), Quebec City, 2015. Google Scholar

[24] Cheng G, Han J, Zhou P. Multi-class geospatial object detection and geographic image classification based on collection of part detectors. ISPRS J Photogrammetry Remote Sens, 2014, 98: 119-132 CrossRef Google Scholar

[25] Cheng G, Han J. A survey on object detection in optical remote sensing images. ISPRS J Photogrammetry Remote Sens, 2016, 117: 11-28 CrossRef ADS arXiv Google Scholar

[26] Cheng G, Zhou P, Han J. Learning Rotation-Invariant Convolutional Neural Networks for Object Detection in VHR Optical Remote Sensing Images. IEEE Trans Geosci Remote Sens, 2016, 54: 7405-7415 CrossRef Google Scholar

[27] Chen S, Wang H, Xu F. Target Classification Using the Deep Convolutional Networks for SAR Images. IEEE Trans Geosci Remote Sens, 2016, 54: 4806-4817 CrossRef Google Scholar

[28] Cho J H, Park C G. Multiple Feature Aggregation Using Convolutional Neural Networks for SAR Image-Based Automatic Target Recognition. IEEE Geosci Remote Sens Lett, 2018, 15: 1882-1886 CrossRef Google Scholar

[29] Huang L, Liu B, Li B. OpenSARShip: A Dataset Dedicated to Sentinel-1 Ship Interpretation. IEEE J Sel Top Appl Earth Observations Remote Sens, 2018, 11: 195-208 CrossRef Google Scholar

[30] Li B, Liu B, Huang L, et al. OpenSARShip 2.0: a large-volume dataset for deeper interpretation of ship targets in Sentinel-1 imagery. In: Proceedings of 2017 SAR in Big Data Era: Models, Methods and Applications (BIGSARDATA), Beijing, 2017. 1--5. Google Scholar

[31] Bao M, Zhang J, Meng J M, et al. Construction and feature analysis of high resolution SAR ship sample set (in Chinese). Chin J Radio Sci, 2019, 34 DOI: 10.13443/j.cjors.2019043007. Google Scholar

[32] Wang Y, Wang C, Zhang H. A SAR Dataset of Ship Detection for Deep Learning under Complex Backgrounds. Remote Sens, 2019, 11: 765 CrossRef Google Scholar

[33] Liu J, Zhang Q. The overview of Gaofen-3 satellite applications. Satellite application, 2018(06) : 12-16. Google Scholar

[34] Song J, Oh K, Kim I, et al. Application of maritime AIS (Automatic Identification System) to ADS-B (Automatic Dependent Surveillance-Broadcast) transceiver. In: Proceedings of International Conference on Control, Automation and Systems, Gyeonggi-do, 2010. 2233--2237. Google Scholar

[35] Pan J, Zheng X, Sun L, et al. Image segmentation based on 2D OTSU and simplified swarm optimization. In: Proceedings of 2016 International Conference on Machine Learning and Cybernetics (ICMLC), 2016. 1026--1030. Google Scholar

[36] Hou X, Ao W, Xu F. End-to-end automatic ship detection and recognition in high-resolution Gaofen-3 spaceborne SAR images. In: Proceedings of IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2019. 9486--9489. Google Scholar

[37] Miller M L, Stone H S, Cox I J. Optimizing Murty's ranked assignment method. IEEE Trans Aerosp Electron Syst, 1997, 33: 851-862 CrossRef ADS Google Scholar

[38] Munkres J. Algorithms for the Assignment and Transportation Problems. J Soc Industrial Appl Math, 1957, 5: 32-38 CrossRef Google Scholar

[39] Yue J, Wang S. EM algorithm and its initialization in Gaussian mixture model based clustering. Microcomput Inf, 2006, 22: 244--246. Google Scholar

[40] He H, Lin Y, Chen F. Inshore Ship Detection in Remote Sensing Images via Weighted Pose Voting. IEEE Trans Geosci Remote Sens, 2017, 55: 3091-3107 CrossRef Google Scholar

[41] Xu F, Wang H, Jin Y Q. Deep learning as applied in SAR target recognition and terrain classification (in Chinese). J Radars, 2017, 6: 136--148. Google Scholar

  • Figure 1

    (Color online) (a) GF-3 example scene and (b) the corresponding aerial image.

  • Figure 2

    (Color online) Distribution of ships in the Huangpu River of Shanghai at 20:56 on November 18, 2018 as extracted from AIS dataset.

  • Figure 3

    (Color online) The framework of SAR-AIS matchup and ship annotation.

  • Figure 4

    (Color online) The flowchart of the multi-scale CFAR ship detection algorithm.

  • Figure 5

    (Color online) AIS-SAR co-registration.

  • Figure 6

    (Color online) Result of AIS-SAR co-registration. (a) SAR image and (b) AIS reports co-registered with SAR image.

  • Figure 7

    (Color online) SAR-AIS matchup via Hungarian algorithm. (a) Original SAR image; (b) SAR detected ships matched-up with AIS reports.

  • Figure 8

    (Color online) The flowchart of sea-land segmentation.

  • Figure 9

    (Color online) The experimental results of LGMM. (a) Bimodal-peak vs. (b) single-peak.

  • Figure 10

    (Color online) The results of sea-land segmentation.

  • Figure 11

    (Color online) Ship detection result of multi-scale CFAR algorithm (detailed information is shown in Table S1). (a) Before and (b) after false alarm discrimination.

  • Figure 12

    FUSAR-Ship of marine objects. From top to bottom, they are ships, strong scatterers, bridges, coastal lands & islands, sea clutter waves, random sea, and land sample patches, respectively.

  • Figure 13

    The structure of 7-category CNN.

  • Figure 14

    (Color online) The results of one example image (from left to right, top to bottom). (a) GF-3 SAR images, (b) AIS-SAR co-registration, (c) sea-land segmentation, (d) ship detection, (e) false-alarm discrimination, (f) ship target contour extraction, (g) final result of ship detection, (h) local amplification result of (g)-1, (i) local amplification result of (g)-2.

  • Figure 15

    (Color online) The taxonomy of SAR ships.

  • Figure 16

    Different categories of ships in FUSAR-Ship.

  • Figure 17

    (Color online) The distribution of FUSAR-Ship datasets of ships. From top to bottom, they are bulk carriers, general cargos, containers, other cargos, fishing, tankers, other ships, and false alarms, respectively.

  • Table 1   GF-3 technical parameters as compared with Sentinel-1 and RadarSat-2
    GF-3 Sentinel-1 RadarSat-2
    Frequency (km) 755 693 798
    Peak power (kW) 1.5 4.7 1.27
    Incident angle $10^{\circ}$–$60^{\circ}$ $20^{\circ}$–$45^{\circ}$ $10^{\circ}$–$60^{\circ}$
    Antenna size 15 m$\times$1.5 m 12.3 m$\times$0.84 m 15 m$\times$1.37 m
    Bandwidth (MHz) 240 100 100
    Polarization mode Single/double/quad polarization Single/double polarization Single/double/quad polarization
    Elevation sweep angle $\pm20^{\circ}$ $\pm11^{\circ}$ $\pm20^{\circ}$
    No. of imaging mode Spotlight Mode, Strip Mode, Sweep Mode, Wave Mode, Ultrafine Mode, etc Strip Mode, TOPS Mode, Wave Mode, Ultra-width Mode, etc Strip Mode, Sweep Mode, Ultrafine Mode, etc
    Resolution (m) 0.5–500 5–20 1–100
    Imaging swath (km) 10–650 20–400 20–500
    Lifespan (year) 8 7.5 7.5
  • Table 2   Maritime targets of the FUSAR-Ship dataset
    Ships Strong Bridges & Coastal lands Sea clutter Sea patches Land
    scattererscoastlines& islandswavespatches
    Training data 1296 229 1023 707 1377 1250 1137
    Validation data 555 128 438 303 590 535 487
    Dataset 1851 427 1461 1010 1967 1785 1624
  • Table 3   The confusion matrix of marine objects classification
    Category Ships Strong Bridges & Coastal lands Sea clutter Random sea Random
    scattererscoastlines& islandswavesland
    Ships 529 15 0 0 1 0 0
    Strong scatterers 14 103 0 0 0 0 2
    Bridges & coastlines 3 1 428 3 1 0 0
    Coastal lands & islands 4 3 6 287 4 1 3
    Sea clutter waves 2 6 0 3 575 2 0
    Random sea 0 0 1 1 4 528 13
    Random land 3 0 3 9 5 2 469
    Sum 555 128 438 303 590 535 487
    Accuracy (%) 95.32 80.47 97.72 94.72 97.48 98.69 96.30
    Overall accuracy = 96.15%
  • Table 4   The example of AIS ship information
    Attribute CHANGKUN7 HONGFAN6 SHUIWU XINHAIHUA ZHESHENGYU07817 HANGONGJIAO4
    CallSign BHNR2 0 BINW 0 0 0
    IMO 0 0 0 413441210 0 0
    MMSI 413363550 413784469 413370630 413441210 900307817 413824589
    ShipTypeEN Tanker Cargo ship LawEnforceVessel Passenger ship WIG OtherTypeOfShip
    NavStatusEN Under way Under way Moored Under way Unknown Under way
    using engineusing engineusing engine using engine
    Length 104 78 30 38 50 25
    Width 15 14 4 7 6 7
    Draught 5.2 3.9 1.6 1.5 0 0
    Heading 511 249 511 298 511 178.2
    Course 241.4 249.6 293.4 80 268 16.1
    Speed 8.9 5.4 0 0 1.9 0.1
    Lon117.40E 117.36E 121.46E 122.18E 123.18E 118.3E
    Lat 30.47N 30.46N 31.27N 29.56N 31.20N 31.11N
    Dest AN QING SHANGHAI SHENJIAM
    UnixTime 1487200911 1487221804 1487086914 1487047384 1487080221 1487250424
    Lon_d 117.66848 117.605323 121.771713 122.302167 123.303615 118.06088
    Lat_d 30.792795 30.772788 31.466342 29.944667 31.339332 31.193668
    pos_y 1861 10527 6100 6143 20705 1707
    pos_x 13030 4070 20916 20144 21913 6004
    ship_y 1850 10530 6094 6104 16636 1614
    ship_x 13040 4084 20814 20164 21440 6000
    Polarization mode DH DH DH DV DV DV
    Incident angle 27.28 27.28 27.28 40.36 40.36 40.36
    Resolution 1.726$\times$1.124 1.726$\times$1.124 1.726$\times$1.124 1.736$\times$1.124 1.736$\times$1.124 1.736$\times$1.124
  • Table 5   The confusion matrix of marine objects classification for ship discrimination
    Category Bulk General Container Other False Fishing Other Tanker
    carrier cargocargoalarmship
    Bulk carrier 240 0 0 0 0 0 2 1
    General cargo 0 120 0 5 1 5 25 16
    Container 0 0 65 4 12 4 9 0
    Other cargo 10 5 3 166 18 3 37 7
    False alarm 0 1 5 11 2406 8 21 2
    Fishing 0 1 0 1 1 10 3 0
    Other ship 82 51 1 116 102 76 519 35
    Tanker 0 0 0 1 2 1 0 52
    Accuracy (%) 72.28 67.42 87.84 54.62 4.65 9.35 84.25 46.18

Copyright 2020  CHINA SCIENCE PUBLISHING & MEDIA LTD.  中国科技出版传媒股份有限公司  版权所有

京ICP备14028887号-23       京公网安备11010102003388号