Research on Video SAR Moving Target Detection Algorithm Based on Improved Faster Region-based CNN
-
摘要: 针对传统视频SAR(ViSAR)运动目标检测方法存在的帧间配准难度大、快速运动目标阴影特征不明显、虚警概率高等问题，该文提出一种基于改进快速区域卷积神经网络(Faster R-CNN)的视频SAR运动目标检测方法。该方法结合Faster R-CNN深度学习算法，利用K-means聚类方法对anchor box的长宽及长宽比进行预处理，并采用特征金字塔网络(FPN)架构对视频SAR运动目标的“亮线”特征进行检测。与传统方法相比，该方法具有实现简单、检测概率高、虚警概率低等优势。最后，通过课题组研制的Mini-SAR系统获取的实测视频SAR数据验证了新方法的有效性。
-
关键词:
- 视频SAR /
- 运动目标检测 /
- 快速区域卷积神经网络 /
- 特征金字塔网络 /
- K-means
Abstract: To solve the problems of inter-frame registration difficult, unclear shadow characteristics of fast moving targets and high false alarm probability in traditional Video Synthetic Aperture Radar (ViSAR) moving target detection methods, a novel video SAR moving target detection method based on improved Faster Region-based Convolutional Neural Networks (Faster R-CNN) is proposed. Combining with the deep learning algorithm of Faster R-CNN, the new method applies the K-means clustering method to preprocess the length, width and aspect ratio of the anchor box. Besides, the Feature Pyramid Networks (FPN) network architecture is used to detect the ‘bright line’ feature of the video SAR moving targets. Compared with traditional methods, the proposed method has the advantages of simple implementation, high detection probability and low false alarm probability. Finally, the effectiveness of the new method is verified by the measured video SAR data obtained from the Mini-SAR system developed by our project team. -
表 1 部分MiniSAR系统参数
雷达参数 具体参数 雷达体制 MCW, DECHIRP 工作频率 8.8～10.6 GHz 工作带宽 1.8 GHz\900 MHz\450 MHz 结构尺寸 150 mm×150 mm×60 mm 重量 ≤3 kg 
下载: 导出CSV
表 2 有无FPN 结构的性能对比
模型 准确率 漏检率 误检率 Faster R-CNN 0.590 0.410 0.108 结合FPN的Faster R-CNN 0.819 0.181 0.084
下载: 导出CSV
表 3 具体参数对训练结果的影响
不同参数 训练2w步 训练4w步 训练5w步 Loss 准确率 Loss 准确率 Loss 准确率 LR=0.010 ReLU 0.643 0.728 0.621 0.811 0.616 0.812 Leaky-ReLU 0.599 0.735 0.574 0.817 0.574 0.817 LR=0.001 ReLU 0.636 0.793 0.582 0.853 0.577 0.855 Leaky-ReLU 0.473 0.824 0.435 0.871 0.435 0.872
下载: 导出CSV
表 4 anchor box设置对训练结果的影响
anchor box 训练2w步 训练4w步 训练5w步 Loss 准确率 Loss 准确率 Loss 准确率 经验值 0.574 0.807 0.556 0.843 0.556 0.845 K-means聚类所得 0.477 0.824 0.433 0.871 0.433 0.872
下载: 导出CSV
表 5 不同模型下运动目标检测性能对比
模型 准确率 漏检率 误检率 Faster R-CNN 0.774 0.226 0.083 Faster R-CNN+FPN 0.845 0.155 0.054 Faster R-CNN+FPN+K-means 0.872 0.128 0.047
下载: 导出CSV
-
杨磊, 李埔丞, 李慧娟, 等. 稳健高效通用SAR图像稀疏特征增强算法[J]. 电子与信息学报, 2019, 41(12): 2826–2835. doi: 10.11999/JEIT190173YANG Lei, LI Pucheng, LI Huijuan, et al. Robust and efficient sparse-feature enhancementfor generalized SAR imagery[J]. Journal of Electronics &Information Technology, 2019, 41(12): 2826–2835. doi: 10.11999/JEIT190173 邓云凯, 禹卫东, 张衡, 等. 未来星载SAR技术发展趋势[J]. 雷达学报, 2020, 9(1): 1–33. doi: 10.12000/JR20008DENG Yunkai, YU Weidong, ZHANG Heng, et al. Forthcoming spaceborne SAR development[J]. Journal of Radars, 2020, 9(1): 1–33. doi: 10.12000/JR20008 肖忠源, 张振华, 马晓萌. 视频SAR发展现状研究[C]. 第三届航天电子战略研究论坛论文集(遥测遥控专刊), 北京, 2017: 37–40.XIAO Zhongyuan, ZHANG Zhenhua, and MA Xiaomeng. Research on the development status of video SAR[C]. Aerospace Electronics Strategic Research Forum, Beijing, China, 2017: 37–40. 苟立婷, 李勇, 朱岱寅, 等. 基于GPU的圆迹视频SAR实时成像算法[J]. 雷达科学与技术, 2019, 17(5): 550–556, 563. doi: 10.3969/j.issn.1672-2337.2019.05.013GOU Liting, LI Yong, ZHU Daiyin, et al. A real-time algorithm for circular video SAR imaging based on GPU[J]. Radar Science and Technology, 2019, 17(5): 550–556, 563. doi: 10.3969/j.issn.1672-2337.2019.05.013 金林, 吴福伟, 杨予昊, 等. 机载视频合成孔径雷达成像技术研究[J]. 微波学报, 2020, 36(1): 45–48. doi: 10.14183/j.cnki.1005-6122.202001008JIN Lin, WU Fuwei, YANG Yuhao, et al. Study of airborne video synthetic aperture radar[J]. Journal of Microwaves, 2020, 36(1): 45–48. doi: 10.14183/j.cnki.1005-6122.202001008 张营, 朱岱寅, 俞翔, 等. 一种VideoSAR动目标阴影检测方法[J]. 电子与信息学报, 2017, 39(9): 2197–2202. doi: 10.11999/JEIT161394ZHANG Ying, ZHU Daiyin, YU Xiang, et al. Approach to moving targets shadow detection for VideoSAR[J]. Journal of Electronics &Information Technology, 2017, 39(9): 2197–2202. doi: 10.11999/JEIT161394 DAMINI A, MANTLE V, and DAVIDSON G. A new approach to coherent change detection in VideoSAR imagery using stack averaged coherence[C]. 2013 IEEE Radar Conference (RadarCon13), Ottawa, Canada, 2013: 1–5. 丁金闪. 视频SAR成像与动目标阴影检测技术[J]. 雷达学报, 2020, 9(2): 321–334. doi: 10.12000/JR20018DING Jinshan. Focusing algorithms and moving target detection based on Video SAR[J]. Journal of Radars, 2020, 9(2): 321–334. doi: 10.12000/JR20018 韩冬, 周良将, 焦泽坤, 等. 基于改进三维后向投影的多圈圆迹SAR相干三维成像方法[J]. 电子与信息学报, 2021, 43(1): 131–137. doi: 10.11999/JEIT190945HAN Dong, ZHOU Liangjiang, JIAO Zekun, et al. A coherent 3-D imaging method for multi-circular SAR based on an improved 3-D back projection algorithm[J]. Journal of Electronics &Information Technology, 2021, 43(1): 131–137. doi: 10.11999/JEIT190945 朱岱寅, 张营, 俞翔, 等. 微型合成孔径雷达成像信号处理技术[J]. 雷达学报, 2019, 8(6): 793–803. doi: 10.12000/JR19094ZHU Daiyin, ZHANG Ying, YU Xiang, et al. Imaging signal processing technology for miniature synthetic aperture radar[J]. Journal of Radars, 2019, 8(6): 793–803. doi: 10.12000/JR19094 张文明, 姚振飞, 高雅昆, 等. 一种平衡准确性以及高效性的显著性目标检测深度卷积网络模型[J]. 电子与信息学报, 2020, 42(5): 1201–1208. doi: 10.11999/JEIT190229ZHANG Wenming, YAO Zhenfei, GAO Yakun, et al. A deep convolutional network for saliency object detection with balanced accuracy and high efficiency[J]. Journal of Electronics &Information Technology, 2020, 42(5): 1201–1208. doi: 10.11999/JEIT190229 HOSANG J, BENENSON R, DOLLÁR P, et al. What makes for effective detection proposals?[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 38(4): 814–830. doi: 10.1109/TPAMI.2015.2465908 ZHAO Liang, LI Chujun, WU Xiaodong, et al. Improved damage characteristics identification method of concrete CT images based on region convolutional neural network[J]. International Journal of Pattern Recognition and Artificial Intelligence, 2020, 34(6): 2054018. doi: 10.1142/S021800142054018X 陶新民, 徐晶, 杨立标, 等. 一种改进的粒子群和K均值混合聚类算法[J]. 电子与信息学报, 2010, 32(1): 92–97. doi: 10.3724/SP.J.1146.2008.01698TAO Xinmin, XU Jing, YANG Libiao, et al. Improved cluster algorithm based on K-means and particle swarm optimization[J]. Journal of Electronics &Information Technology, 2010, 32(1): 92–97. doi: 10.3724/SP.J.1146.2008.01698 侯志强, 刘晓义, 余旺盛, 等. 基于双阈值-非极大值抑制的Faster R-CNN改进算法[J]. 光电工程, 2019, 46(12): 190159. doi: 10.12086/oee.2019.190159HOU Zhiqiang, LIU Xiaoyi, YU Wangsheng, et al. Improved algorithm of Faster R-CNN based on double threshold-non-maximum suppression[J]. Opto-Electronic Engineering, 2019, 46(12): 190159. doi: 10.12086/oee.2019.190159 -
图(8) / 表(5)
计量
- 文章访问数: 1994
- HTML全文浏览量: 741
- PDF下载量: 203
- 被引次数: 0
下载:
