Incremental learning approach based on vector neural network for emitter identification

To deal with the problem of emitter identification (EID) caused by the measurement uncertainty of emitter feature parameters and to realise the automatic updating of the emitter database, which is usually used as emitter templates in identification processing, a vector neural network based incremental learning (VNNIL) approach for EID is proposed. This method combines the vector neural networks (VNNs) and the ensemble-based incremental learning (Learn++) algorithm. The VNN is adopted to construct a weak classifier and the Learn++ is used to generate ensembles of the weak classifiers. Considering that the VNN can realise the non-linear mapping between the interval-value input data and the interval-value output emitter types, and that the Learn++ can update the emitter database automatically, the VNNIL treats the two mentioned problems above as a single one and realises EID and parameters updating at the same time. A number of simulations are presented to demonstrate the identification and updating capability of the VNNIL algorithm. As shown in the simulation results, the VNNIL algorithm not only possesses a better learning and identification capability, but also achieves a better noise adaptability.     

Approach based on combination of vector neural networks for emitter identification

To deal with the problem of emitter identification caused by the measurement uncertainty of emitter feature parameters, this study proposes a new identification algorithm based on combination of vector neural networks (CVNN), which is deduced from the backpropagation vector neural network and can realise the nonlinear mapping between the interval-value input data and the interval-value output emitter types. The key idea of CVNN is to adopt a combination of multiple multi-input/single-output neural networks to construct an identification system; each of the networks can only realise the identification function between two emitter types. Through quantitative analysis, it can be concluded that the proposed algorithm requires less computational load in the training stage. A number of simulations are presented to demonstrate the identification capability of the CVNN algorithm for emitter signals with and without additive noise. Simulation results show that the proposed algorithm not only has better identification capability, but also is relatively more insensitive to noise.     

基于对角切片特征提取和深度学习的辐射源识别

针对复杂电磁环境下辐射源识别率低的问题，提出基于对角切片特征和深度学习的辐射源识别算法。利用辐射源信号双谱的个体特性，提取信号双谱对角切片特征作为深度学习模型的输入数据，采用Softmax分类器进行辐射源识别。仿真实验利用两部同型辐射源进行测试，结果表明该算法能识别个体辐射源，在低信噪比条件下也能获得高的辐射源识别率；相比于其他识别算法，双谱对角切片特征有更鲁棒的分辨性。     

Multi-Parameter Signal Sorting Algorithm Based on Dynamic Distance Clustering

A multi-parameter signal sorting algorithm for interleaved radar pulses in dense emitter environment is presented. The algorithm includes two parts, pulse classification and pulse repetition interval （PRI） analysis. Firstly, we propose the dynamic distance clustering （DDC） for classification. In the clustering algorithm, the multi-dimension features of radar pulse are used for reliable classification. The similarity threshold estimation method in DDC is derived, which contributes to the efficiency of the algorithm. However, DDC has large computation with many signal pulses. Then, in order to sort radar signals in real time, the improved DDC （IDDC） algorithm is proposed. Finally, PRI analysis is adopted to complete the process of sorting. The simulation experiments and hardware implementations show both algorithms are effective.     

Multi-Parameter Signal Sorting Algorithm Based on Dynamic Distance Clustering

A multi-parameter signal sorting algorithm for interleaved radar pulses in dense emitter environment is presented. The algorithm includes two parts, pulse classification and pulse repetition interval (PRI) analysis. Firstly, we propose the dynamic distance clustering (DDC) for classification. In the clustering algorithm, the multi-dimension features of radar pulse are used for reliable classification. The similarity threshold estimation method in DDC is derived, which contributes to the efficiency of the algorithm. However, DDC has large computation with many signal pulses. Then, in order to sort radar signals in real time, the improved DDC (IDDC) algorithm is proposed. Finally, PRI analysis is adopted to complete the process of sorting. The simulation experiments and hardware implementations show both algorithms are effective.     

雷达辐射源信号脉内特征分析

脉内特征提取是新型雷达辐射源信号识别的关键问题。本文提出一种新的雷达辐射源信号脉内特征提取和识别方法．将雷达辐射源脉冲信号的分形维数作为识别脉内调制方式的分类特征,这些特征包含了雷达辐射源信号幅度、频率和相位等的变化和分布信息,反映了雷达辐射源信号脉内调制规律,理论分析和仿真实验结果都证明了这些特征具有对噪声不敏感的良好特性．通过10种典型雷达辐射源信号的特征提取和分类识别的实验结果表明,本文所提取的脉内特征类间距离大、类内距离小、正确识别率高．证实了本文方法的有效性。     

基于脉冲相位线性度的雷达辐射源信号分选

为解决信号分选中出现的“漏批”问题,提高信号分选准确率,提出了一种基于脉冲相位线性度的雷达辐射源信号分选新方法.对脉冲相位线性度的检测过程进行推导,并通过仿真实验对推导结果加以验证.提取分选特征参数,给出了信号分选算法的步骤,并进行分选仿真分析.仿真实验表明,本方法可以在较低输入信噪比下实现高效准确的雷达辐射源信号分选,在电子情报侦察系统上有着广阔的应用前景.     

基于扩张残差网络的雷达辐射源信号识别

针对低信噪比条件下,复杂多类雷达辐射源信号识别存在特征提取困难,识别正确率低的问题,本文提出了一种基于时频分析和扩张残差网络的辐射源信号自动识别方法.首先通过时频分析将信号时域波形转换成二维时频图像以反映信号本质特征;然后进行时频图像预处理以保留时频图像完备信息,适应深度学习模型输入;最后构建扩张残差网络以自动提取信号时频图像特征,实现雷达辐射源信号分类识别.实验结果表明,信噪比为-6dB时,该方法对16类雷达辐射源信号的整体识别正确率能够达到98.2%,对时频图像特征相似的类LFM（Linear Frequency Modulation）信号的整体识别正确率超过95%.本文提供了一种新的雷达辐射源信号智能识别方法,具有较好的工程应用前景.     

基于时差多参分选的多层感知器网络脉间识别

现代战争中,雷达系统发展迅速。为识别复杂的雷达信号调制模式以及混合体制雷达,该文提出一种基于多站获取脉冲时差参数联合其他脉冲描述字分选的办法,利用多层感知器神经网络得到脉间调制识别结果。该文通过时差参数与其他脉冲描述字去交错解决传统脉冲重复周期估计算法无法对复杂的脉间调制方式进行估计。利用训练好的多层感知器,获取完成去交错后的脉冲序列其特征向量,获得其脉间调制类型识别。通过实验仿真,在脉冲丢失率不高于20%情况下,对复杂脉间调制方式的正确识别概率在90%以上。     

基于栈式自编码机和模糊函数主脊的雷达辐射源信号识别

针对人工提取雷达辐射源信号特征存在提取周期长、特征描述不完备等局限性,提出了一种基于深度学习栈式自编码机和模糊函数主脊的雷达信号识别方法.该方法根据信号模糊函数主脊包含丰富的内在调制信息的特点,从信号中提取用于分类识别的抽象特征.通过对六种雷达辐射源信号进行实验,并对比人工特征提取及其他深度学习方法,结果表明,本文所提方法在信噪比（signal-noise ratio,SNR）为2 dB以上时均能保持100%的识别准确率,SNR为-6 dB时识别准确率仍能保持82.83%以上,明显高于其他方法.即使在包含相同调制类型不同参数的信号环境中,当SNR大于0 dB时识别率均稳定在95.0%以上,SNR降低到-4 dB时识别率也能达到79.0%.证明该方法能有效提取到信号的深层特征,且具有良好的抗噪性能,基本满足实际战场的需求.     

基于同步压缩小波变换的主信号抑制技术

在辐射源个体识别(SEI)技术中,能量较高的主信号往往导致微弱个体特征稳定性降低,进而影响最终的个体识别效果。为了解决该问题并提升辐射源个体识别性能,该文提出基于同步压缩小波变换的主信号抑制技术。首先,利用静态小波变换完成对带噪信号的去噪预处理;然后,利用同步压缩小波变换完成对主信号的检测和抑制,并以均方根误差和皮尔逊相关系数为数值指标,验证算法的有效性;最后,在主信号抑制的基础上,利用分形理论中盒维数完成对信号的特征提取,并利用单核支持向量机验证个体识别性能。实验结果表明,与主信号抑制之前相比,主信号抑制算法下个体识别率提升了10%左右,验证了同步压缩小波变换的主信号抑制算法对辐射源个体识别率提升的有效性。     

基于脉间特征深度学习的雷达辐射源识别

多功能雷达在复杂程序调度下,发射信号参数呈现取值范围宽、捷变速度快、变化随机性强等特点,非合作接收方难以对其建立有效的信号模型,给电子侦察系统的雷达辐射源识别带来严峻挑战。本文提出一种基于深度学习的复杂体制雷达辐射源识别方法,利用大样本全脉冲数据形成脉间参数变化的图像特征表示,从宏观上揭示雷达辐射源隐含的波形设计机理,并设计了基于AlexNet网络的图像特征深度学习网络开展辐射源识别,实测数据实验表明了本文的方法对一定时间跨度内的有限部同型多功能雷达具有良好的识别性能,为多功能雷达辐射源智能个体识别提供了新的解决思路。      

基于改进深度卷积神经网络的纸币识别研究

针对如何提高纸币识别率的问题,该文提出一种改进深度卷积神经网络(DCNN)的纸币识别算法。该算法首先通过融合迁移学习、带泄露整流(Leaky ReLU)函数、批量归一化(BN)和多层次残差单元构造深度卷积层,对输入的不同尺寸纸币进行稳定而快速的特征提取与学习;然后采用改进的多层次空间金字塔池化算法对提取的纸币特征实现固定大小的输出表示;最后通过网络全连接层和softmax层实现纸币图像分类。实验结果表明,该算法在分类性能、泛化能力与稳定性上明显优于常用的纸币分类算法;同时该算法也能够满足纸币清分系统的实时性要求。     

基于回归分析理论的辐射源个体识别技术

针对目前辐射源个体识别未能将信号特征与硬件组成相联系的问题，该文使用高阶谱分析和变分模态分解(VMD)两种特征提取手段，进行研究分析，采用围线双谱积分以及改进变分模态分解技术对半实物平台仿真信号以及软件仿真(ADS)输出信号进行特征提取并分析。通过软件仿真定量分析辐射源相位噪声以及功率放大电路非线性失真对信号无意调制特征的影响，对变量进行相关性分析，并对其中显著相关的变量进行回归拟合，得到其相关回归函数。然后利用硬件与特征的相关性，改进传统支持向量机(SVM)分类器，构建相关性权重支持向量机分类器。最后分别以软件仿真输出信号以及半实物仿真平台实测信号为样本进行验证，结果表明，同信噪比下权重支持向量机与传统支持向量机相比分类准确率提升在10%以上。     

基于改进的半监督主动学习的雷达信号识别

为解决在雷达信号分类识别过程中训练样本较少的问题,本文提出了联合主动学习和半监督学习,并对其伪标记样本进行迭代验证改进的分类算法。针对复杂的电磁环境下雷达信号识别率低的问题,本文将径向高斯核时频分析应用于雷达信号,并对时频分布进行奇异值分解,提取出奇异向量作为雷达信号识别的特征参数。针对传统的半监督主动学习算法的不足,利用改进的半监督主动学习算法构建分类器,该算法通过对伪标记样本进行迭代验证来提高伪标记信息的准确性,从而改善了最终的分类性能,实现了在可获取的有标签样本数量较少的条件下对雷达信号的高概率识别。仿真结果表明,本文提出的特征识别方法可以获得较高的识别率。      

一种自适应容差的雷达信号分选聚类算法

雷达信号分选的目的就是从交错的、密集复杂的脉冲信号流中提取出同一辐射源的脉冲序列。战场环境中信号流的密集性,信号形式的复杂性,给信号分选带来了严重的挑战。面对如此复杂的信号环境,传统的基于直方图统计的雷达信号分选算法的分选结果可信度越来越差。在聚类雷达信号分选算法的基础之上提出了一种自适应容差的雷达信号聚类算法,克服了传统的雷达信号聚类分选算法中容差选择困难的问题。仿真结果表明该方法能够准确地分选出各个辐射源的脉冲序列。     

A Modified Error Function to Improve the Error Back-Propagation Algorithm for Multi-Layer Perceptrons

This paper proposes a modified error function to improve the error back-propagation (EBP) algorithm for multi-Layer perceptrons (MLPs) which suffers from slow learning speed. It can also suppress overspecialization for training patterns that occurs in an algorithm based on a cross-entropy cost function which markedly reduces learning time. In the similar way as the cross-entropy function, our new function accelerates the learning speed of the EBP algorithm by allowing the output node of the MLP to generate a strong error signal when the output node is far from the desired value. Moreover, it prevents the overspecialization of learning for training patterns by letting the output node, whose value is close to the desired value, generate a weak error signal. In a simulation study to classify handwritten digits in the CEDAR [1] database, the proposed method attained 100% correct classification for the training patterns after only 50 sweeps of learning, while the original EBP attained only 98.8% after 500 sweeps. Also, our method shows mean-squared error of 0.627 for the test patterns, which is superior to the error 0.667 in the cross-entropy method. These results demonstrate that our new method excels others in learning speed as well as in generalization.     

Recognition of Electromyographic Signals Using Cascaded Kernel Learning Machine

Electromyographic (EMG) signals recognition is a complex pattern recognition problem due to its property of large variations in signals and features. This paper proposes a novel EMG classifier called cascaded kernel learning machine (CKLM) to achieve the goal of high-accuracy EMG recognition. First, the EMG signals are acquired by three surface electrodes placed on three different muscles. Second, EMG features are extracted by autoregressive model (ARM) and EMG histogram. After the feature extraction, the CKLM is performed to classify the features. CKLM is composed of two different kinds of kernel learning machines: generalized discriminant analysis (GDA) algorithm and support vector machine (SVM). By using GDA, both the goals of the dimensionality reduction of input features and the selection of discriminating features, named kernel FisherEMG, can be reached. Then, SVM combined with one-against-one strategy is executed to classify the kernel FisherEMG. By cascading SVM with GDA, the input features will be nonlinearly mapped twice by radial-basis function (RBF). As a result, a linear optimal separating hyperplane can be found with the largest margin of separation between each pair of postures' classes in the implicit dot product feature space. In addition, we develop a digital signal processor (DSP)-based EMG classification system for the control of a multi-degrees-of-freedom prosthetic hand for the practical implementation. Based on the clinical experiments, the results show that the proposed CKLM is superior to other frequently used methods, such as k-nearest neighbor algorithm, multilayer neural network, and SVM. The best EMG recognition rate 93.54% is obtained by CKLM.