针对深度学习模型易出现灾难性遗忘的关键难点,提出了一种基于原型对比的合成孔径雷达(synthetic aperture radar,SAR)图像增量小样本目标检测算法-InFSAR(prototype contrast based incremental few-shot SAR object detection)。首先...针对深度学习模型易出现灾难性遗忘的关键难点,提出了一种基于原型对比的合成孔径雷达(synthetic aperture radar,SAR)图像增量小样本目标检测算法-InFSAR(prototype contrast based incremental few-shot SAR object detection)。首先,采用基础数据集对检测器进行预训练,以构建初步的特征提取能力;其次,设计一种类原型表征生成模块,以构建一组能够代表数据内在特征的类原型。在增量学习阶段,设计一种混合类原型对比编码模块,以有效学习新类别与基础类别之间的区分性特征。此外,为缓解灾难性遗忘问题,引入类原型校准策略,使模型在类原型上的预测分布逐步逼近真实分布,从而保持对基础类别识别的稳定性。在小样本目标检测数据集SRSDD-v1.0上的实验结果表明,在5-shot设置下,InFSAR对船舶细粒度目标的检测精度达到46.5%。同时,该方法能够在无需访问基础类训练数据的情况下,实现对少量标注新类别的增量检测与识别。展开更多
Two rupture zones of Wenchuan earthquake have been successfully located by an offset-tracking procedure of ENVISAT ASAR amplitude images.Accuracy of offset-tracking strongly depends on the quality of image coregistrat...Two rupture zones of Wenchuan earthquake have been successfully located by an offset-tracking procedure of ENVISAT ASAR amplitude images.Accuracy of offset-tracking strongly depends on the quality of image coregistration.In order to remove noise caused by coregistration,a combined filter method is proposed to remove noise of estimated offsets.In addition,Kriging interpolation is applied to smooth the azimuth displacement offset map.The displacement offset maps of azimuth and range show that about 245 km and 60 km in length along Beichuan-Yingxiu fault and Guanxian-Jiangyou fault were ruptured by Wenchuan earthquake,respectively.The result of offset-tracking is validated by field observations of China Earthquake Administration.展开更多
随着数字化技术和雷达系统的发展,针对合成孔径雷达(Synthetic Aperture Radar,SAR)的干扰对抗技术不断进步,尤其是基于数字射频存储技术(Digital Radio Frequency Memory,DRFM)产生的有源欺骗干扰为SAR成像系统带来了前所未有的考验。...随着数字化技术和雷达系统的发展,针对合成孔径雷达(Synthetic Aperture Radar,SAR)的干扰对抗技术不断进步,尤其是基于数字射频存储技术(Digital Radio Frequency Memory,DRFM)产生的有源欺骗干扰为SAR成像系统带来了前所未有的考验。针对欺骗干扰开展SAR成像抗干扰方法研究,本文基于相位编码波形与带有循环前缀的正交频分复用(Cyclic Prefix Orthogonal Frequency Division Multiplexing,CP-OFDM)波形进行正交波形设计,提出了相位编码CP-OFDM正交波形。基于CP-OFDM波形的循环前缀(Cyclic Prefix,CP)特征,引入基于线性模型的脉冲压缩方法对相位编码CP-OFDM正交波形的SAR成像回波进行距离向处理,能够实现无旁瓣干扰的自相关脉冲压缩。通过对相位编码CP-OFDM波形的时域相位进行编码优化设计,可以实现不同相位编码CP-OFDM波形之间良好的互相关性能。基于线性模型脉冲压缩方法改善了一种p范数多波形加权循环(p-norm Weighted Cyclic Algorithm,p-WeCAN)波形优化算法,采用该算法对相位编码CP-OFDM波形集的相位编码序列进行优化设计,优化后波形的互相关脉冲压缩结果的峰值水平(Peak Sidelobe Level,PSL)相比于随机相位编码CP-OFDM波形的互相关PSL改善了2 dB左右。CP特性赋予了相位编码CP-OFDM波形良好的自相关脉冲压缩结果,相位编码优化设计提供了良好的互相关脉冲压缩结果,采用该正交波形集进行SAR成像,能够实现对欺骗干扰的抑制。进行了点目标、面目标和基于GF-3回波数据反演的半实测数据的抗欺骗干扰SAR成像仿真,与基于线性调频(Linear Frequency Modulation,LFM)波形的欺骗干扰条件下的成像结果进行对比,验证了相位编码CP-OFDM对欺骗干扰的抑制能力。展开更多
文摘针对深度学习模型易出现灾难性遗忘的关键难点,提出了一种基于原型对比的合成孔径雷达(synthetic aperture radar,SAR)图像增量小样本目标检测算法-InFSAR(prototype contrast based incremental few-shot SAR object detection)。首先,采用基础数据集对检测器进行预训练,以构建初步的特征提取能力;其次,设计一种类原型表征生成模块,以构建一组能够代表数据内在特征的类原型。在增量学习阶段,设计一种混合类原型对比编码模块,以有效学习新类别与基础类别之间的区分性特征。此外,为缓解灾难性遗忘问题,引入类原型校准策略,使模型在类原型上的预测分布逐步逼近真实分布,从而保持对基础类别识别的稳定性。在小样本目标检测数据集SRSDD-v1.0上的实验结果表明,在5-shot设置下,InFSAR对船舶细粒度目标的检测精度达到46.5%。同时,该方法能够在无需访问基础类训练数据的情况下,实现对少量标注新类别的增量检测与识别。
基金sponsored by the Special Item of Professional Fund for Basic Scientific Research of Chinese Central-Level Public-Welfare College/Institute from Chinese Finance Department (ZDJ2009-36)National Natural Science Foundation of China (40774023)+1 种基金National 863 Program of China (2006AA12Z150)the project Dragon 2 (5343)
文摘Two rupture zones of Wenchuan earthquake have been successfully located by an offset-tracking procedure of ENVISAT ASAR amplitude images.Accuracy of offset-tracking strongly depends on the quality of image coregistration.In order to remove noise caused by coregistration,a combined filter method is proposed to remove noise of estimated offsets.In addition,Kriging interpolation is applied to smooth the azimuth displacement offset map.The displacement offset maps of azimuth and range show that about 245 km and 60 km in length along Beichuan-Yingxiu fault and Guanxian-Jiangyou fault were ruptured by Wenchuan earthquake,respectively.The result of offset-tracking is validated by field observations of China Earthquake Administration.
文摘随着数字化技术和雷达系统的发展,针对合成孔径雷达(Synthetic Aperture Radar,SAR)的干扰对抗技术不断进步,尤其是基于数字射频存储技术(Digital Radio Frequency Memory,DRFM)产生的有源欺骗干扰为SAR成像系统带来了前所未有的考验。针对欺骗干扰开展SAR成像抗干扰方法研究,本文基于相位编码波形与带有循环前缀的正交频分复用(Cyclic Prefix Orthogonal Frequency Division Multiplexing,CP-OFDM)波形进行正交波形设计,提出了相位编码CP-OFDM正交波形。基于CP-OFDM波形的循环前缀(Cyclic Prefix,CP)特征,引入基于线性模型的脉冲压缩方法对相位编码CP-OFDM正交波形的SAR成像回波进行距离向处理,能够实现无旁瓣干扰的自相关脉冲压缩。通过对相位编码CP-OFDM波形的时域相位进行编码优化设计,可以实现不同相位编码CP-OFDM波形之间良好的互相关性能。基于线性模型脉冲压缩方法改善了一种p范数多波形加权循环(p-norm Weighted Cyclic Algorithm,p-WeCAN)波形优化算法,采用该算法对相位编码CP-OFDM波形集的相位编码序列进行优化设计,优化后波形的互相关脉冲压缩结果的峰值水平(Peak Sidelobe Level,PSL)相比于随机相位编码CP-OFDM波形的互相关PSL改善了2 dB左右。CP特性赋予了相位编码CP-OFDM波形良好的自相关脉冲压缩结果,相位编码优化设计提供了良好的互相关脉冲压缩结果,采用该正交波形集进行SAR成像,能够实现对欺骗干扰的抑制。进行了点目标、面目标和基于GF-3回波数据反演的半实测数据的抗欺骗干扰SAR成像仿真,与基于线性调频(Linear Frequency Modulation,LFM)波形的欺骗干扰条件下的成像结果进行对比,验证了相位编码CP-OFDM对欺骗干扰的抑制能力。
