Research on PCA and KPCA Self-Fusion Based MSTAR SAR Automatic Target Recognition Algorithm

This paper proposes a PCA and KPCA self-fusion based MSTAR SAR automatic target recognition algorithm. This algorithm combines the linear feature extracted from principal component analysis （PCA） and nonlinear feature extracted from kernel principal component analysis （KPCA） respectively, and then utilizes the adaptive feature fusion algorithm which is based on the weighted maximum margin criterion （WMMC） to fuse the features in order to achieve better performance. The linear regression classifier is used in the experiments. The experimental results indicate that the proposed self-fusion algorithm achieves higher recognition rate compared with the traditional PCA and KPCA feature fusion algorithms.

Radar automatic target recognition based on feature extraction for complex HRRP

Radar high-resolution range profile （HRRP） has received intensive attention from the radar automatic target recognition （RATR） community. Usually, since the initial phase of a complex HRRP is strongly sensitive to target position variation, which is referred to as the initial phase sensitivity in this paper, only the amplitude information in the complex HRRP, called the real HRRP in this paper, is used for RATR, whereas the phase information is discarded. However, the remaining phase information except for initial phases in the complex HRRP also contains valuable target discriminant information. This paper proposes a novel feature extraction method for the complex HRRP. The extracted complex feature vector, referred to as the complex feature vector with difference phases, contains the difference phase information between range cells but no initial phase information in the complex HRRR According to the scattering center model, the physical mechanism of the proposed complex feature vector is similar to that of the real HRRP, except for reserving some phase information independent of the initial phase in the complex HRRP. The recognition algorithms, frame-template establishment methods and preprocessing methods used in the real HRRP-based RATR can also be applied to the proposed complex feature vector-based RATR. Moreover, the components in the complex feature vector with difference phases approximate to follow Gaussian distribution, which make it simple to perform the statistical recognition by such complex feature vector. The recognition experiments based on measured data show that the proposed complex feature vector can obtain better recognition performance than the real HRRP if only the cell interval parameters are properly selected.

A feature extraction method for synthetic aperture radar(SAR) automatic target recognition based on maximum interclass distance

Synthetic aperture radar(SAR) automatic target recognition is an important application in SAR.How to extract features has restricted the application of SAR technology seriously.In this paper,a new feature extraction method for SAR automatic target recognition based on maximum interclass distance is proposed,which integrates class and neighborhood information.This method can reinforce discriminative power using maximum interclass distance,so it can improve recognition rate effectively.

A new feature extraction method using the amplitude fluctuation property of target HRRP for radar automatic target recognition

Due to the aspect sensitivity of high-resolution range profile(HRRP),traditional radar HRRP target recognition methods usually use average profile within some target-aspect region as the target-aspect template.Actually,the amplitude fluctuation property of target HRRP also represents some feature information of the target.Based on the scattering center model,a new feature extraction method using the amplitude fluctuation property of target HRRP is proposed in this paper.The weighted HRRP feature extracted by the new method can represent the scatterer distribution in every range cell,thereby it can describe the scattering property of the target better.The experimental results based on measured data show that the new feature extraction method can greatly improve recognition performances.

New statistical model for radar HRRP target recognition

The mixture of factor analyzers （MFA） can accurately describe high resolution range profile （HRRP） statistical charac- teristics. But how to determine the proper number of the models is a problem. This paper develops a variational Bayesian mixture of factor analyzers （VBMFA） model. This procedure can obtain a lower bound on the Bayesian integral using the Jensen＇s inequality. An analytical solution of the Bayesian integral could be obtained by a hypothesis that latent variables in the model are indepen- dent. During computing the parameters of the model, birth-death moves are utilized to determine the optimal number of model au- tomatically. Experimental results for measured data show that the VBMFA method has better recognition performance than FA and MFA method.

Radar high resolution range profile recognition via multi-SV method

For radar high resolution range profile (HRRP) recognition, three aspects are of great importance to improve the performance, i.e. discrimination for outlier, classification for inner and an accurate description for feature space. To tackle these issues, a novel target recognition method is designed, denoted by the multiple support vectors (multi-SV) method. With the proposed method, a special framework is constructed by a treble correlate support vector model to segment the feature space to two regions with the distribution of density, and then the description and classification hyperplane for each region are achieved. Based on the support vector framework, this method needs less memory and computation complexity to fit practical radar HRRP recognition. Finally, the experiment based on the measured data verifies the excellent performance of this method.

THE INFLUENCE OF SPECKLE ON HIGH RESOLUTION RANGE PROFILE RECOGNITION BASED ON THE MATCHING SCORE

Template database is the key to radar automation target recognition based on High Resolution Range Profile (HRRP). From the traditional perspective, average HRRP is a valid template for it can represent each HRRP without scatterer Moving Through Range Cell (MTRC). However, template database based on this assumption is always challenged by measured data. One reason is that speckle happens in the frame without scatterer MTRC. Speckle makes HRRP fluctuate sharply and not match well with the average HRRP. We precisely introduce the formation mechanism of speckle. Then, we make an insight into the principle of matching score. Based on the conclusion, we study the properties of matching score between speckled HRRP and the average HRRP. The theoretical analysis and Monte Carlo experimental results demonstrate that speckle makes HRRP not to match well with the average HRRP according to the energy ratio of speckled scatterers. On the assumption of ideal scattering centre model, speckled HRRP has a matching score less than 85% with the average HRRP if speckled scatterers occupy more than 50% energy of the target.

Radar HRRP statistical recognition with temporal factor analysis by automatic Bayesian Ying-Yang harmony learning

Radar high-resolution range profiles(HRRPs)are typical high-dimensional and interdimension dependently distributed data,the statistical modeling of which is a challenging task for HRRP-based target recognition.Supposing that HRRP samples are independent and jointly Gaussian distributed,a recent work[Du L,Liu H W,Bao Z.IEEE Transactions on Signal Processing,2008,56(5):1931–1944]applied factor analysis(FA)to model HRRP data with a two-phase approach for model selection,which achieved satisfactory recognition performance.The theoretical analysis and experimental results reveal that there exists high temporal correlation among adjacent HRRPs.This paper is thus motivated to model the spatial and temporal structure of HRRP data simultaneously by employing temporal factor analysis(TFA)model.For a limited size of high-dimensional HRRP data,the two-phase approach for parameter learning and model selection suffers from intensive computation burden and deteriorated evaluation.To tackle these problems,this work adopts the Bayesian Ying-Yang(BYY)harmony learning that has automatic model selection ability during parameter learning.Experimental results show stepwise improved recognition and rejection performances from the twophase learning based FA,to the two-phase learning based TFA and to the BYY harmony learning based TFA with automatic model selection.In addition,adding many extra free parameters to the classic FA model and thus becoming even worse in identifiability,the model of a general linear dynamical system is even inferior to the classic FA model.

A WEIGHTED FEATURE REDUCTION METHOD FOR POWER SPECTRA OF RADAR HRRPS

Feature reduction is a key process in pattern recognition. This paper deals with the feature reduction methods for a time-shift invariant feature, power spectrum, in Radar Automatic Target Recognition (RATR) using High-Resolution Range Profiles (HRRPs). Several existing feature reduction methods in pattern recognition are analyzed, and a weighted feature reduction method based on Fisher's Discriminant Ratio (FDR) is proposed in this paper. According to the characteristics of radar HRRP target recognition, this proposed method searches the optimal weight vector for power spectra of HRRPs by means of an iterative algorithm, and thus reduces feature dimensionality. Compared with the method of using raw power spectra and some existing feature reduction methods, the weighted feature reduction method can not only reduce feature dimensionality, but also improve recognition performance with low computation complexity. In the recognition experiments based on measured data, the proposed method is robust to different test data and achieves good recognition results.

Feature Extraction of Radar Range Profiles Based on Normalized Central Moments

The normalized central moments are widely used in pattern recognition because of scale and translation invariance. The moduli of normalized central moments of the 1-dimensional complex range profiles are used here as feature vector for radar target recognition. The common feature extraction method for high resolution range profile obtained by using Fourier-modified direct Mellin transform is inefficient and unsatisfactory in recognition rate And. generally speaking, the automatic target recognition method based on inverse synthetic aperture radar 2-dimensional imaging is not competent for real time object identification task because it needs complicated motion compensation which is sometimes too difficult to carry out. While the method applied here is competent for real-time recognition because of its computational efficiency. The result of processing experimental data indicates that this method is good at recognition.

增强-检测级联SAR地面目标检测网络

在合成孔径雷达地面目标检测任务中,传统检测方法因为在处理过程中采用固定模型假设而导致性能严重下降。卷积神经网络作为一种基于数据驱动的方法,在拥有足够的训练集时可以显著提高目标检测的准确性,但在检测陆地背景下的微小目标时性能仍不稳定。为了应对这些挑战,提出了一种先增强后检测的地面目标检测框架。其中包括以Transformer为骨干网络的增强网络、增强目标特征区分度的跨特征空间注意力模块以及具有多尺度特征的检测网络。形成一个级联的目标检测网络架构,以实现更好的推理性能。使用MSTAR基准数据集对提出的网络进行实验,证明提出的级联网络在各项指标上超过其他现有方法,其精度最高可以达到93.6%。

基于语义引导层次化分类的雷达地面目标HRRP识别方法

高分辨距离像(HRRP)反映了目标空间散射结构在雷达视线方向的投影,近年来被认为是地面目标识别的重要途径。现有的HRRP识别方法采用手工特征加传统机器学习分类器,均属于平面分类方法,即采用统一标准不加区别的优选特征并单次决策最终类别。然而该方法在实际应用中面临种类繁杂、数据不平衡、HRRP姿态敏感性等诸多问题,难以获取最佳的应用效果。层次化方法采取分而治之思想,将一个复杂的细粒度识别任务拆解为多个简单的识别子任务。本文采用层次化识别的思路,提出了一种基于语义引导层次化分类的雷达地面目标识别方法。该方法以联合语义和数据构建的树形结构将一个复杂的细粒度识别任务拆解为多个简单的识别子任务,并针对每一个识别子任务匹配一套优选特征集和一个局部分类器。本方法在仿真数据和实测数据上完成了验证。实验结果表明了本文方法处理地面目标识别任务的有效性。

空中目标动态电磁散射数据仿真系统设计与实现

电磁散射数据是目标识别研究的基础,但由于试验测量成本高、可重复性差等问题,空中目标电磁散射实测数据十分有限。基于去遮挡的N点模型和电磁计算数据插值研究了空中目标动态电磁散射数据仿真方法,设计了空中目标动态电磁数据仿真系统,该系统将电磁散射仿真与实际飞行场景相结合,支持场景自定义和空中目标三维模型库扩展。分析了飞机和巡航导弹2类目标仿真数据及成像结果,结果验证了仿真方法正确性和系统有效性,可为研究实际场景下的空中目标识别提供支撑。

基于GCN和CNN联合的SAR图像自动目标识别

基于卷积神经网络(Convolutional Neural Network, CNN)的合成孔径雷达(Synthetic Aperture Radar,SAR)自动目标识别(Automatic Target Recognition, ATR)技术近些年来备受关注,已成为SAR图像解译领域的研究热点。然而,这类方法主要利用的是SAR图像的幅值信息,仅从局部区域中提取特征。鉴于SAR图像中的目标通常被视为散射中心的相干叠加,这些目标展现出复杂的结构和丰富的上下文信息。仅依靠CNN难以充分捕捉目标周围的全局信息,这可能会影响识别精度。因此,为了进一步提高识别性能,本研究引入图卷积网络(Graph Convolutional Network, GCN),提出一种结合GCN和CNN的SAR ATR方法。该方法首先利用传统CNN提取与SAR图像幅值相关的局部特征,接着通过构造图数据并应用GCN提取全局特征。此外,本研究还设计了多尺度GCN,通过融合不同尺度的特征来增强模型对图数据的学习能力。在模型训练阶段,采用标签平滑技术以缓解过拟合问题。通过端到端的训练策略,实现了GCN和CNN参数的联合优化,从而实现高精度的SAR图像目标识别。最终,通过在MSTAR和OpenSARship数据集上的实验表明,所提方法在识别性能上优于现有技术,并展现出卓越的泛化能力。

面向SAR目标识别成像参数敏感性的深度学习技术研究进展

随着人工智能技术的发展,基于深度神经网络的合成孔径雷达(SAR)目标识别得到了广泛关注。然而,SAR系统的成像机制导致了图像特性与成像参数之间的强相关性,因此深度学习框架下的目标识别算法精度极易受成像参数敏感性的干扰,这成为了制约先进智能算法部署到实际工程中的一大障碍。该文首先回顾了SAR图像目标识别技术的发展与相关数据集,从雷达工作的成像几何、载荷参数和噪声干扰3个角度,深入分析了成像参数变化对图像特性的影响;然后,从模型、数据、特征3个维度,总结归纳了现有文献关于深度学习技术对成像参数敏感性的鲁棒性与泛化性这一问题的研究进展;接下来,汇总并分析了典型方法的实验结果;最后讨论了在未来有望突破成像参数敏感性这一问题的深度学习技术研究方向。

基于自适应阈值卷积网络的抗干扰雷达目标识别

本文提出了一种自适应阈值卷积网络(ATCN),基于HRRP数据进行抗干扰雷达目标识别。ATCN中的核心模块是自适应阈值卷积单元(ATCU),该模块能准确高效地完成对HRRP数据的特征提取。在ATCU中,采用自适应阈值函数充当激活函数,自动调整阈值以面对不同信干比的数据;利用多个不同尺度的卷积核来捕获HRRP数据中的区域差异特征;引入通道注意力机制和残差连接优化网络结构。本文进行了大量的抗干扰目标识别实验,实验结果表明,相比于所选择的3种对比网络,本文的ATCN网络能在不同干扰类型和不同信干比下提供更优的平均识别率和更好的指标稳定性,且具有更少的网络模型参数量和浮点运算次数,具备轻量化和高效的特点。

基于多域特征融合的HRRP目标识别方法

高分辨一维距离像(High Resolution Range Profile,HRRP)包含了丰富的目标信息,通过提取HRRP的强散射点分布特征可实现对不同目标的分类和识别。论文充分考虑了HRRP的多域特征和时间依赖性,利用ResNet18网络进行时频域特征提取,并结合记忆融合网络(Memory Fusion Network,MFN)提出新型深度学习模型MI-MFN(Multi Input-MFN)进行多域特征的融合识别,实现了在不同维度的记忆的跨视图交互,有效地学习和提取HRRP序列特征。实验结果表明,MI-MFN模型的识别准确率可以达到99.9%以上,具有出色的识别性能。

SAR-ATR系统复数对抗样本生成方法

针对现有对抗攻击方法只能用于攻击实数卷积神经网络这一限制,提出了一种基于生成对抗网络的复数对抗样本生成方法。首先,设计了一种产生有效对抗样本的复数模型,并引入了复数计算模块;其次,利用残差神经网络作为基本骨架,将预训练的复数网络作为判别器实现对抗训练,以增强对抗样本的攻击能力;最后,通过替代模型实现可迁移的对抗攻击,以此实现了更高的攻击成功率。实验结果表明,所提方法在有目标攻击和无目标攻击任务下的成功率分别达到了76.338%和87.841%,迁移的成功率更高且对抗样本与原始干净样本更为接近。所提方法将对抗攻击扩展到复数神经网络后,避免了合成孔径雷达目标信息和精度的丢失,为实际合成孔径雷达自动目标识别系统的安全性和鲁棒性提供了参考方案。

小样本SAR目标的双重一致性因果识别方法

在小样本条件下提升方法的泛化性能,是合成孔径雷达自动目标识别(SAR ATR)的重要研究方向。针对该方向中的基础理论问题,该文建立了一个SAR ATR因果模型,证明了SAR图像中背景、相干斑等干扰在充足样本条件下可以被忽略;但在小样本条件下,这些因素将成为识别中的混杂因子,在提取的SAR图像特征中引入虚假相关性,影响SAR ATR性能。为了甄别和消除这些特征中的虚假效应,该文提出一个基于双重一致性的小样本SAR ATR方法,其中双重一致性包括类内一致性掩码和效应一致性损失。首先,基于鉴别特征应具有类内一致和类间差异的原则,利用类内一致性掩码,捕获目标的类内一致鉴别特征,甄别出目标特征中的混淆部分,准确估计出干扰引入的虚假效应。其次,基于不变风险最小化的思想,利用效应一致性损失,将经验风险最小化数据量需求转变为对效应相似度的度量需求,降低虚假效应消除对数据量的需求,消除特征中的虚假效应。因而,所提基于双重一致性的小样本SAR ATR方法可实现特征提取中的真实因果,实现准确的识别性能。两个基准数据集上的识别实验,验证了该方法的合理性和有效性,可提升小样本条件下SAR目标识别的性能。

基于三维电磁散射参数化模型的SAR目标识别方法

合成孔径雷达目标识别是雷达数据解译中一个长期研究的难点问题。近年来,基于模型的SAR目标识别方法由于在扩展条件下的识别性能表现良好而备受关注。在联合国内多家研究单位进行攻关的基础上,该文简要阐述了对该问题的初步研究成果及思考。首先从3个方面出发梳理了散射部件模型发展的技术脉络并对其进行了补充完善;然后从正向推算和逆向反演两条技术途径提出了复杂目标电磁散射参数化建模方法;最后提出了基于复杂目标电磁散射参数化模型的目标识别新框架。论文最后对基于模型的SAR目标识别下一步研究方向进行了展望。