基于运动与表象特征的广域船舶目标识别方法

严荣慧; 谢海成; 花敏恒; 羊箭锋

doi:10.19693/j.issn.1673-3185.02320

基于运动与表象特征的广域船舶目标识别方法

doi: 10.19693/j.issn.1673-3185.02320

1.
苏州大学文正学院, 江苏苏州 215104
2.
苏州大学电子信息学院, 江苏苏州 215006

基金项目: 2021江苏省高等学校自然科学研究面上项目（21KJD510006）；苏州大学文正学院2020年高等教育改革研究课题（2910342520）; 2018江苏省高等学校自然科学研究面上项目（18KJD510009）

详细信息

作者简介:
严荣慧，女，1993年生，硕士，讲师。研究方向：模式识别与人工智能。E-mail: wzj029@suda.edu.cn

谢海成，男，1990年生，科研助理。研究方向：信号与信息处理。E-mail: 14919380932@qq.com

花敏恒，男，1995年生，硕士生。研究方向：图像处理。E-mail: 773605595@qq.com

羊箭锋，男，1978年生，博士，高级实验师。研究方向：信号与信息处理。E-mail: jfyang@suda.edu.cn

通信作者:
羊箭锋

中图分类号: U665.2；TN958.98
计量
- 文章访问数: 692
- HTML全文浏览量: 169
- PDF下载量: 86
- 被引次数: 0
出版历程
- 收稿日期: 2021-03-18
- 修回日期: 2021-06-20
- 网络出版日期: 2022-02-22
- 刊出日期: 2022-03-02

Wide-area ship target recognition method based on motion and appearance features

1.
Wenzheng College, Soochow University, Suzhou 215104, China
2.
School of Electronic and Information Engineering, Soochow University, Suzhou 215006, China

基于运动与表象特征的广域船舶目标识别方法由严荣慧，等创作，采用知识共享署名4.0国际许可协议进行许可。

摘要

摘要: 目的旨在提出对航行于关键广阔水域内的船舶进行准确识别和定位的改进方法。方法运用视频监控的优点，综合采用基于背景差分算法的运动目标检测方法与基于深度学习算法的图像表象特征识别方法，结合目标的运动特征和图像表象特征，实现多维度广域船舶识别的功能，并对水纹降噪、多级运动检测、航道监控图像窗口分割检测等方法进行改进，进一步提高航行监控系统的船舶识别准确率。结果现场航道监控验证结果表明，采用所提改进方法可以准确识别航道监控画面中任意类型和尺度的船舶，且使用常规摄像头即可实现半径3 km范围内的船舶识别、定位效果。结论所提方法具有监控范围广、船舶类型全覆盖、自动目标识别、抗干扰能力强等优点。
- 广域航道监控 /
- 船舶识别 /
- 运动检测 /
- 多维特征 /
- YOLO检测算法
Abstract: Objective The aim of this paper is to proposes methods for better recognizing and positioning ships sailing in critical and wide-area waterways during monitoring operation. Methods Based on video surveillance technology, the joint use of the motion and appearance features of ship target is carried out to realize a wide-area multi-dimensional recognition function via the combination of background subtraction based moving object detection algorithm and deep learning based target recognition algorithm. In addition, the improved approaches including water ripple noise reduction, hierarchical moving object detection and window segmentation of waterway monitoring image are put forward to further improve recognition accuracy. Results The field demonstration results show that the improved methods proposed in this paper allow the accurate recognition of a ship of any type or size on the monitoring screen, and the use of conventional cameras can also achieve the recognition and position of a ship navigating a water area within a radius of 3 km. Conclusions The improved methods proposed in this study have a range of advantages including wide-area monitoring, complete coverage of ship types and sizes, automatic target recognition and robust anti-interference ability.
- wide-area waterway monitoring /
- ship recognition /
- motion detection /
- multi-dimensional features /
- YOLO detection algorithm

HTML全文

图 1 航道监控中远、近水域内船舶目标对比

Figure 1. Comparison of ship target in the far and near water areas during waterway monitoring

下载: 全尺寸图片幻灯片

图 2 基于背景差分算法对小目标船舶的识别效果

Figure 2. Recognition effect of small-sized ship target by background subtraction algorithm

下载: 全尺寸图片幻灯片

图 3 基于背景差分算法对运动目标检测效果

Figure 3. Detection effect of motion target by background subtraction algorithm

下载: 全尺寸图片幻灯片

图 4 渐变滤波示意图

Figure 4. Schematic diagram of gradual filtering

下载: 全尺寸图片幻灯片

图 5 水纹噪声特征

Figure 5. Features of water ripple noise

下载: 全尺寸图片幻灯片

图 6 航道中目标面积占比与目标距离的关系

Figure 6. The relationship between targets area ratio and target distance

下载: 全尺寸图片幻灯片

图 7 水纹滤波效果图

Figure 7. Effect images of water ripple filtering

下载: 全尺寸图片幻灯片

图 8 快慢二级运动目标检测结果

Figure 8. Results of two-level detection for the fast and slow motion targets

下载: 全尺寸图片幻灯片

图 9 中间带水域内小目标船舶检测图像

Figure 9. Image of detected small-sized ship target in the intermediate waterways

下载: 全尺寸图片幻灯片

图 10 航道监控画面中的窗口分割

Figure 10. Window segmentation of waterway monitoring image

下载: 全尺寸图片幻灯片

图 11 基于YOLO检测算法的船舶目标识别效果

Figure 11. Effect of ship target recognition based on YOLO detection algorithm

下载: 全尺寸图片幻灯片

图 12 基于二维特征的船舶目标识别流程图

Figure 12. Process of ship target recognition based on two-dimensional features

下载: 全尺寸图片幻灯片

图 13 基于二维特征的船舶目标识别效果

Figure 13. Effect of ship target recognition based on two-dimensional features

下载: 全尺寸图片幻灯片

表 1 船舶目标识别原算法与其改进方法检测效果对比

Table 1. Effect comparison of ship target recognition by original and improved algorithms

方法	识别准确率/%	识别召回率/%	最远识别距离/km	能否识别类型
背景差分算法	<0.01	97.8	3	否
YOLO检测算法	95.1	54.7	0.2	能
基于水纹滤波和多级检测的背景差分目标识别	68.4	98.4	3	否
基于窗口分割的 YOLO目标识别	97.2	68.5	0.4	能
基于改进窗口分割和 YOLO目标识别	97.2	98.4	3	能

下载: 导出CSV

参考文献(16)

[1]	吴飞, 李志特. 新时期中国内河航运发展问题分析[J]. 珠江水运, 2020(15): 87–88. WU F, LI Z T. Analysis of China's inland waterway development in the New Era[J]. Pearl River Water Transport, 2020(15): 87–88 (in Chinese).
[2]	罗本成, 张姝慧. 新形势新要求驱动内河航运发展步入“快车道”[N]. 中国水运报, 2020-08-17(002). LUO B C, ZHANG S H. The new situation and new requirements drive the development of inland waterway into the "fast lane"[N]. China Water Transport News, 2020-08-17(002) (in Chinese).
[3]	许正路. 5米以下游船水上旅游活动安全监管研究[J]. 中国海事, 2014(9): 49–52. doi: 10.3969/j.issn.1673-2278.2014.09.021 XU Z L. Research on the water sightseeing safety supervision of sightseeing boat below 5 meters in scenic spots[J]. China Maritime Safety, 2014(9): 49–52 (in Chinese). doi: 10.3969/j.issn.1673-2278.2014.09.021
[4]	李峰, 易宏. 无人水面艇在水上交通安全监管中的应用[J]. 中国舰船研究, 2018, 13(6): 27–33. LI F, YI H. Application of USV to maritime safety supervision[J]. Chinese Journal of Ship Research, 2018, 13(6): 27–33 (in Chinese).
[5]	严新平, 王树武, 马枫. 智能货运船舶研究现状与发展思考[J]. 中国舰船研究, 2021, 16(1): 1–6. YAN X P, WANG S W, MA F. Review and prospect for intelligent cargo ships[J]. Chinese Journal of Ship Research, 2021, 16(1): 1–6 (in Chinese).
[6]	杨英. 基于视频图像的船舶流量检测方法研究[D]. 武汉: 武汉理工大学, 2019. YANG Y. Research on ship flow detection method based on video image[D]. Wuhan: Wuhan University of Technology, 2019 (in Chinese).
[7]	XU C, YIN C J, WANG D Z, et al. Fast ship detection combining visual saliency and a cascade CNN in SAR images[J]. IET Radar, Sonar & Navigation, 2020, 14(12): 1879–1887.
[8]	CORDOVA A W A, QUISPE W C, INCA R J C, et al. New approaches and tools for ship detection in optical satellite imagery[J]. Journal of Physics, Conference Series, 2020, 1642(1): 012003.
[9]	CAO C Q, WU J, ZENG X D, et al. Research on airplane and ship detection of aerial remote sensing images based on convolutional neural network[J]. Sensors, 2020, 20(17): 4696. doi: 10.3390/s20174696
[10]	马吉顺, 吴天明, 韩鹏, 等. 基于YOLO 检测算法的船舶识别定位系统[J]. 新型工业化, 2019, 9(9): 33–37. MA J S, WU T M, HAN P, et al. Ship identification and positioning system based on YOLO algorithm[J]. The Journal of New Industrialization, 2019, 9(9): 33–37 (in Chinese).
[11]	高岚. 基于视频的多运动目标检测算法研究[D]. 沈阳: 沈阳理工大学, 2008. GAO L. Research on multiple moving objects detection algorithm based on the video[D]. Shenyang: Shenyang Ligong University, 2008 (in Chinese).
[12]	乐英, 赵志成. 基于背景差分法的多运动目标检测与分割[J]. 中国工程机械学报, 2020, 18(4): 305–309. LE Y, ZHAO Z C. Multi-moving object detection and segmentation based on background difference method[J]. Chinese Journal of Construction Machinery, 2020, 18(4): 305–309 (in Chinese).
[13]	魏俊杰. 水岸图像的中值滤波和分割算法的研究及实现[D]. 长春: 吉林大学, 2011. WEI J J. On study and realization of median filtering and segmentation algorithm of waterfront images[D]. Changchun: Jilin University, 2011 (in Chinese).
[14]	李冬琴, 王丽铮, 王呈方. 支持向量机回归方法在船型要素建模中的应用[J]. 中国舰船研究, 2007, 2(3): 18–21,39. doi: 10.3969/j.issn.1673-3185.2007.03.004 LI D Q, WANG L Z, WANG C F. Method of support vector regression in modeling ship principal particulars [J]. Chinese Journal of Ship Research, 2007, 2(3): 18 –21,39 (in Chinese). doi: 10.3969/j.issn.1673-3185.2007.03.004
[15]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, NV, USA: IEEE, 2016: 779-788.
[16]	LU S Y, LIU J H, WANG B Z, et al. A novel object detection algorithm in video[C]//Proceedings of 2019 the 9th International Workshop on Computer Science and Engineering (WCSE 2019). Hong Kong, China:Science and Engineering Institute (SCIEI), 2019:8.