基于随机矩阵最大特征值分布的频谱感知算法

现有频谱感知算法在低信噪比时检测性能较低且受虚警概率影响大,针对此问题,提出了一种基于wishart矩阵样本协方差矩阵最大特征值的分布特性的频谱感知算法。该算法利用最大特征值与几何平均特征值的比值,不需要主用户的先验知识,不敏感于噪声,对相关信号和独立同分布信号均具有较高的检测性能。仿真结果表明,所提算法受虚警概率的影响较小,检测性能高,并且在采样点数、协作用户数、信噪比及虚警概率较小的情况下,也能获得较好的检测性能。     

Quadratic covariance matrix-based cooperative spectrum sensing method by using an evolutionary algorithm

Cognitive radio (CR) is a practical technology to solve the current low utilization of spectrum resources, and spectrum sensing is the most critical technique in a CR network. In this paper, a genetic simulated annealing algorithm based on quadratic covariance matrix and information geometry is proposed for cooperative spectrum sensing (CSS) to enhance the performance in the low signal-noise ratio (SNR). Firstly, the quadratic covariance matrix of cooperative secondary users (SUs) is used as the characteristic matrix to perform feature extraction. Secondly, based on the information geometry, the characteristic matrix is mapped on the statistical manifold to avoid information loss. Furthermore, the genetic simulated annealing algorithm is used to obtain a classifier on the statistical manifold, and the mutation process is improved by a new mutation operator to accelerate the convergence speed of the whole algorithm. Finally, the classifier is employed to implement spectrum sensing. In the simulation analysis, the proposed method has better spectrum sensing performance than the popular various methods under low SNR and faster convergence speed.     

MASS-RAB: Robust adaptive beamforming for general-rank signal models via matched spatial spectrum processing

The wavefront of acoustic signal suffers from fast fluctuation after a long distance propagation in a random and inhomogeneous ocean channel, which makes the rank of the covariance matrix for the desired signal (signal of interest) remarkably higher than one. Consequently, the assumption of rank-one point signal model for existing adaptive beamforming algorithms is no longer suitable. In this paper, a matched spatial spectrum processing based robust adaptive beamforming (MASS-RAB) algorithm is presented for general-rank signal models. First, the interference-plus-noise covariance matrix and the desired signal covariance matrix are reconstructed using the matched spatial spectrum processing method. Second, the weight vector is directly calculated using these reconstructed covariance matrices for the minimum variance distortionless response (MVDR) algorithm, which is developed for the general-rank signal models. Due to covariance matrix reconstruction, the MASS-RAB algorithm is more robust than those methods relying on the sample covariance matrix. The cases of the rank-one point signal model and the full-rank non-point signal model are considered by several numerical examples. Experimental results have demonstrated the superiority of the proposed MASS-RAB method.     

基于置信学习机与近红外光谱的煤种快速分类方法

基于近红外光漫反射谱技术的检测分析具有简单,快捷,安全等优势而被广泛应用于各行各业。应用近红外光谱分析技术实现不同煤种的快速分类,该方法可以替代费时费力费财的传统化学分析方法。同时首次将置信学习机(confidence machine)引入近红外分析中,实现了对分析结果的风险评估。采集了来自不同矿区共四种不同煤种(肥煤,焦煤,瘦煤和贫瘦煤)的199个煤样本的近红外光谱,通过机器学习的方法针对煤的近红外光谱构建了煤种分类器来实现煤种的快速分类。在近红外分析中引入了置信学习机的分析方式,结合支持向量机(SVM),构建了离线和在线的CM-SVM分类器。置信学习机是一种概率方法,使用概率(CM-SVM)来取代分类超平面(SVM)进行分类,不仅分类效果好于传统的SVM,达到了95.48%的分类率,还能同时给出每个样本分类结果的置信度,可靠度等风险信息。另外,CM-SVM通过对置信水平的设定,得到不同置信度下预测区间,该区间的预测正确率是与置信水平严格对应的,对于产品质量控制有非常重要的意义。置信学习机同时是一种在线的学习模型,新样本的不断加入会提高模型的性能,非常适合于工业现场的在线分析。在线的CM-SVM模型随着样本数的增加,预测结果的置信度有所提高,对工业现场近红外分析有重要意义。     

基于矢量水听器阵列空域预滤波处理的多重信号分类算法研究

矢量水听器能同时拾取声压和振速信息,在相同的信噪比、阵元数及阵列孔径下,矢量阵定向性能优于声压阵列。目前,以多重信号分类算法(Multiple signal classification,MUSIC)为代表的高分辨定向算法已经广泛应用于矢量水听器阵列中。但是随着信噪比降低、信号源方位间隔减小,传统MUSIC算法定向精度及分辨概率显著下降。本文采用最小二乘法设计适用于矢量水听器水平阵列的矩阵空域滤波器,用于阵列数据的空间滤波预处理,可以对阻带扇面噪声进行有效抑制。由滤波后的数据协方差矩阵可以得到新的噪声子空间,在传统MUSIC算法基础上修正通带扇面内阵列流型的畸变后即可得到滤波后MUSIC算法的方位谱。仿真结果表明,当信噪比较低时,改进算法有效提高了通带扇面内目标方位分辨性能。最后本文对四基元矢量水平阵列海试数据进行了处理,改进算法对窄带信号定向较常规算法-3 dB束宽减小了13°,旁瓣级降低约8 dB。对有一定带宽的行船辐射噪声定向处理得到了更加精确的航迹图,海试数据处理结果证明了该算法的可行性和有效性。     

Secure collaborative cognitive radio based on chaotic modulation and compressive sensing

Cognitive radio (CR) is a wireless technology that is used to overcome the spectrum scarcity problem. CR includes several stages, spectrum sensing is the first stage in the CR cycle. Traditional spectrum sensing (SS) techniques have many challenges in the wideband spectrum. CR security is an important problem, since when an attacker from outside the network access the sensing information this produces an increase in sensing time and reduces the opportunities for exploiting vacant band. Compressive sensing (CS) is proposed to capture all the wideband spectrum at the same time to solve the challenges and improve the performance in the traditional techniques and then one of the traditional SS techniques are applied to the reconstructed signal for detection purpose. The sensing matrix is the core of CS must be designed in a way that produces a low reconstruction error with high compression. There are many types of sensing matrices, the chaotic matrix is the best type in terms of security, memory storage, and system performance. Few works in the literature use the chaotic matrix in CS based CR and these works have many challenges: they used sample distance in the chaotic map to generate a chaotic sequence which consumes high resources, they did not take into consideration the security in reporting channel, and they did not measure their works using real primary user (PU) signal of a practical application under fading channel and low SNR values. In this paper, we propose a chaotic CS based collaborative scenario to solve all challenges that have been presented. We proposed a chaotic matrix based on the Henon map and use the differential chaotic shift keying (DCSK) modulation to transmit the measurement vector through the reporting channel to increase the security and improve the performance under fading channel. The simulation results are tested based on a recorded real-TV signal as PU and Compressive Sampling Matching Pursuit (CoSaMP) recovery algorithm under AWGN and TDL-C fading channels in collaborative and non-collaborative scenarios. The performance of the proposed system has been measured using recovery error, mean square error (MSE), derived probability of detection (Pdrec), and sensitivity to initial values. To measure the improvement introduced by the proposed system, it is evaluated in comparison with selected chaotic and random matrices. The results show that the proposed system provides low recovery error, MSE, with high Pdrec, security, and compression under SNR equal to −30 dB in AWGN and TDL-C fading channels as compared to other matrices in the literature.     

基于克隆选择支持向量机高光谱遥感影像分类技术

作为支持向量机(support vector machine, SVM)高光谱影像分类的一个重要环节,参数设置的效率和精度直接影响到SVM模型训练效率和最终分类精度。本文首先建立一个SVM高光谱影像分类器,提出了利用免疫克隆选择算法优化的交叉验证进行核函数参数和惩罚因子C的优化选择的方法,得到了一种基于克隆选择优化的支持向量机(clonal selection SVM, CSSVM)高光谱影像分类器。然后将CSSVM与传统的基于网格搜索交叉验证的支持向量机（gird search SVM, GSSVM）分类器进行了对比评价,评价指标包括模型训练时间和分类精度等。最后基于AVIRIS高光谱遥感影像进行了两算法分类对比试验,结果表明：提出的CSSVM测试样本总分类精度超过85.1%和Kappa系数超过0.821 3,影像总分类精度超过81.58%和Kappa系数超过0.772 8,CSSVM与GSSVM的分类精度差别在0.08%以内,Kappa系数差别在0.001以内;CSSVM的模型训练时间是GSSVM的1/6至1/10,得到显著缩短;CSSVM方法在保持传统GSSVM优良分类精度的基础上,极大提高了模型的训练效率。     

ISG-FHEV等效燃油消耗最小控制策略

针对室内复杂环境下火灾识别准确率会降低的问题,提出了一种改进的粒子群算法优化支持向量机参数进行火灾火焰识别的方法。首先在 颜色空间进行火焰图像分割,对获得的火焰图像进行预处理并提取相关特征量;其次采用PSO算法搜索SVM的最优核参数和惩罚因子,并在PSO算法中加入变异操作和非线性动态调整惯性权值的方法,加快了搜索SVM最优参数的精度和速度;然后将提取的火焰各个特征量作为训练样本输入SVM模型进行训练,并建立参数优化后的SVM分类器模型;最后将待测试样本输入SVM模型进行分类识别。算法的火灾识别准确率达到94.09%,分类效果明显优于其他分类算法。仿真结果表明,改进的PSO优化SVM算法提高了火焰识别的准确率和实时性,算法的自适应性更强,误判率更低。     

Spectrum Sensing Implemented with Improved Fluctuation-Based Dispersion Entropy and Machine Learning

Spectrum sensing is an important function in radio frequency spectrum management and cognitive radio networks. Spectrum sensing is used by one wireless system (e.g., a secondary user) to detect the presence of a wireless service with higher priority (e.g., a primary user) with which it has to coexist in the radio frequency spectrum. If the wireless signal is detected, the second user system releases the given frequency to maintain the principle of not interfering. This paper proposes a machine learning implementation of spectrum sensing using the entropy measure as a feature vector. In the training phase, the information about the activity of the wireless service with higher priority is gathered, and the model is formed. In the classification phase, the wireless system compares the current sensing report to the created model to calculate the posterior probability and classify the sensing report into either the presence or absence of wireless service with higher priority. This paper proposes the novel application of the Fluctuation Dispersion Entropy (FDE) measure recently introduced in the research community as a feature vector to build the model and implement the classification. An improved implementation of the FDE (IFDE) is used to enhance the robustness to noise. IFDE is further enhanced with an adaptive method (AIFDE) to automatically select the hyper-parameter introduced in IFDE. Then, this paper combines the machine learning approach with the entropy measure approach, which are both recent developments in spectrum sensing research. The approach is compared to similar approaches in literature and the classical energy detection method using a generated radar signal data set with different conditions of SNR(dB) and fading conditions. The results show that the proposed approach is able to outperform the approaches from literature based on other entropy measures or the Energy Detector (ED) in a consistent way across different levels of SNR and fading conditions.     

可见/近红外光谱结合GWO-SVM对千禧番茄的分类鉴别

千禧番茄营养丰富且酸甜可口,不同千禧番茄品种的风味和营养价值均有明显差异,尤其是番茄红素、柠檬酸、维生素C和氨基酸含量的差异较大。传统人工分类方式效率低、主观性强、误检率高等问题亟待解决。为筛选综合营养价值高且风味佳的千禧番茄品种,实现千禧番茄快速准确分类,提出了基于千禧番茄光谱特征的分类模型构建及GWO优化SVM算法研究,以期解决千禧番茄自动化分类问题。以四个品种千禧番茄作为研究对象,试验样本240个,将其按2∶1比例划分为训练集160个和测试集80个样本,利用可见-近红外光谱采集系统获取350～1 000 nm范围内的千禧番茄反射强度,经光谱校正得样本反射率;为增强信噪比,截取481.15～800.03 nm范围内的光谱波段作为有效波段。由于数据采集过程受无关信息干扰影响建模效果,故将平滑点数设置为3进行Savitzky-Golay(SG)平滑预处理。预处理后采用连续投影算法(SPA)提取特征波长变量,优选得到11个特征波长反射率作为输入矩阵X,预设样本变量1, 2, 3和4作为输出矩阵Y,利用支持向量机(SVM)建立SPA-SVM千禧番茄定性分类模型,训练集和测试集平均分类准确率...     

基于多分类支持向量机的船舶桨叶数识别研究

分析了目前常用的支持向量机多分类方法以及存在的不足,本文提出了一种混合纠错输出编码的多分类支持向量机改进算法,并应用于利用船舶目标辐射噪声DEMON谱进行船舶桨叶数分类的实验。理论分析与实验结果表明,该改进算法编码明确、具备纠错能力,是一种有效的多分类支持向量机方法,在船舶桨叶数识别中,其分类性能优于一对余、一对一及最小输出编码支持向量机等多分类方法,可适用于船舶桨叶数的分类识别。     

混沌信号的压缩感知去噪

对非线性时间序列进行噪声抑制是从中提取有效信息的前提. 混沌信号的去噪算法不仅要使滤波后的信号具有较高的信噪比, 也要具有较好的不确定性. 从压缩感知的角度出发,提出了一种新的噪声抑制方法. 该方法包括估计噪声方差, 以及依据动态的稀疏度将观测值往确定的过完备字典上投影. 仿真实验表明, 该方法比常用的小波阈值法和局部曲线拟合法具有更高的输出信噪比, 而原始信号的混沌特性也能得到较大程度的恢复.     

基于LVQ与SVM算法的近红外光谱煤产地鉴别

传统煤产地鉴别方法一般以发热量、挥发分、粘结指数、哈氏可磨指数和坩埚膨胀序数作为分类指标,过程复杂耗时较多、耗费巨大的人力、物力并且无法直接快速的得到煤样产地等问题,借助近红外光谱技术快速无损检测的优势,利用基于SVM的留一算法对光谱数据集进行异常样本剔除,得到包含正确光谱信息的煤样光谱数据集,构造基于SVM算法与LVQ算法的定性分析模型,完成基于近红外光谱分析技术的煤产地的快速鉴别,无需对煤样的各种指标进行汇总并且人为预测。针对SVM分析模型中存在随机参数优化问题,引入PSO算法对SVM模型中的损失参数C和核函数半径g进行改进,得到最优参数,最后引入计算准确率的方法对比以上模型并进行评价分析。实验一共收集了加拿大、俄罗斯、澳大利亚、印度尼西亚、中国内蒙等5个地区的煤样光谱数据集,数据集共计305组煤样样本,其中异常样本共计10组,分别选择各国煤炭光谱的前31组作为训练样本,后6组数据作为测试样本,结果表明各分类模型的分类准确率均能达到75%以上,其中基于PSO算法改进的SVM分析模型的准确率可达到96.67%,仅一个样本出现问题,可快速高效地实现基于近红外光谱分析技术的煤产地的鉴别。     

光谱分析中的支持向量机方法及其性能优化

针对红外光谱气体分析中建立数据模型需要标定大量样本的问题, 提出一种基于正则理论的支持向量机的小样本机器学习方法,该方法能在获得模型参数全局最优点的同时保证训练误差为零,因而能较好地消除光谱间的交叉敏感现象,利用其良好的非线性映射能力对多组分红外光谱仪的试验结果表明,该方法可使光谱仪的交叉灵敏度下降约81倍。针对支持向量机(SVM)没有足够的理论支持的结构参数选取比较困难的问题,提出一种基于遗传算法和交叉检验相结合的遗传支持向量机(GA_SVM)算法,利用遗传算法的随机搜索特性求取SVM的最优结构参数,在20世代即可求取光谱仪的最小均方根误差(MSE)0.018, 并且在算法的前数世代,系统的MSE即已开始成倍下降。这些结果表明GA_SVM光谱仪具有更高的效率和泛华能力。     

应用半正定规划的目标方位超分辨方法

针对水下目标方位超分辨估计问题,提出了一种基于半正定规划(Sdp)的常规波束(CBF)方位超分辨算法(SdpCBF).Sdp-CBF算法基于常规波束形成获得多目标方位谱数据,利用阵列响应矩阵和半正定规划技术,精确估计目标数量和波达角方向.该算法的本质是利用阵列特性和信号能量信息获得超分辨方位估计,不用进行子空间分解,通过卷积反演的方式将阵列孔径的有限效应消除,在L₂范数约束条件下重构空间谱.仿真表明,Sdp-CBF算法具有较强的噪声抑制能力,对非相干和相干信号均具有目标方位超分辨能力,在低信噪比环境下的方位分辨性能超过多重信号分类(MUSIC)等经典高分辨算法。对消声水池以及湖上实验数据的处理结果显示,Sdp-CBF算法在复杂环境中对相干信号及微弱信号具有较强的分辨能力。      

A novel THz spectroscopy recognition method for transgenic organisms based on APSO combined with SVM

Currently, the transgenic products detection methods are mostly based on visible/near-infrared light spectrum. In addition, it is hard to set up the parameters in the support vector machine (SVM) model and there is a large amount of calculation on spectrum data. To solve these problems, this paper proposed an algorithm based on terahertz (THz) spectrum and SVM using adaptive particle swarm optimize (APSO-SVM) for building up the classifications of transgenic cotton seed. To conduct the transgenic cotton seed classification, within the wavelength region 150 μm—3 mm, the THz spectrums are first sampled from 165 samples of three newest transgenic cotton seeds. Then, the 165 transgenic cotton seeds are recognized based on the APSO-SVM. Experiment results indicate that the total recognition rate is up to 97.3%, which prove that the THz spectrum combined with APSO-SVM can provide a reliable, rapid, simple and nondestructive detection method for transgenic cotton seed.     

基于时频谱特征的白酒品质分类方法研究

为了建立快速、准确的白酒品质鉴别方法,利用机器学习方法对不同品质的白酒建模。为了提取不同品质白酒的特征,使用离子迁移谱对不同品质白酒进行分析,构建了基于白酒离子迁移谱信号的特征向量,并对不同品质的白酒进行了识别与分类。白酒样本的离子迁移谱信号通过利用美国Excellims公司GA2100型电喷雾-离子迁移谱仪（ESI-IMS）采集获得,每一个离子迁移谱信号是强度随时间变化的时间序列信号;提取了原始数据离子迁移谱的时域特征谱峰。为了获得更全面的特征,对离子迁移谱数据进行了傅里叶变换并提取频域内的特征谱峰。同时为了表述信号变化的特征,计算了离子迁移谱的谱熵和过零率,构建N×9维的特征向量矩阵;使用主成分分析(PCA)和线性判别分析(LDA)分别对上述获得的特征进行了特征降维,其中使用PCA对特征向量矩阵降维后的前三维特征对整体特征的累计贡献率达到了95%,而使用LDA对特征向量矩阵降维后的前两维特征对整体特征的累计贡献率就达到了95%。因此,选择了LDA作为特征降维方法;最后,利用机器学习中的非线性分类器支持向量机(SVM)对白酒离子迁移谱数据进行分类研究。实验结果表明,在真酒和添加酒精的白酒二分类中,SVM方法正确分类率达到100%;而在真酒和分别添加10%,20%,30%,40%和50%酒精浓度的五种假酒的六分类中,SVM方法正确分类率达到99.7%。比较了逻辑回归(LRM)分类、模糊C均值分类(FCM)和K近邻分类(KNN)对白酒样本离子迁移谱分类实验结果。研究表明,对于离子迁移谱非常接近的真酒和添加酒精的白酒,基于频谱特征向量的SVM方法能够准确的区分开来,为白酒的品质鉴别提供了一种新的检测方法。     

基于太赫兹衰减全反射光谱的水质分析

随着人口的增长和社会的迅速发展,水资源短缺和水污染问题日益严重。水质分类作为水质污染评估工作中的一项重要环节,其意义和作用也更加突出。基于太赫兹衰减全反射（THz-ATR）光谱和模式识别技术,提出了一种水质分析模型。利用太赫兹时域光谱系统和衰减全反射模块测量了纯净水、自来水、河水、海水A和海水B五种水样的太赫兹衰减全反射光谱,通过光学参数提取模型获得0.2～1.0 THz频率范围内五种水样的折射率、吸收系数、介电常数实部和介电常数虚部。利用主成分分析（PCA）对折射率进行降维和特征提取,分别作出样品在第一、二主成分上的二维得分图和前三个主成分上的三维得分图,结果显示,基于折射率的主成分得分图可以明显的区分不同的水样。为了进一步对不同水样进行准确分类,将降维之后的数据输入到支持向量机（SVM）中构建水样分类模型,每种水样随机选取其中的五分之三作为训练集,剩余的数据作为测试集,同时引入网格搜索（GridSearch）、遗传算法（GA）和粒子群（PSO）三种优化算法对支持向量机参数进行优化。结果显示,基于网格搜索算法的支持向量机最优参数c和g分别为1.414 2和2.0,准确率为99.0%;基于遗传算法的支持向量机最优参数c和g分别为1.675 4和5.966 5,准确率为99.5%;基于粒子群算法的支持向量机最优参数c和g分别为3.154 9和12.589,准确率为100%。可以看出,使用不同的优化算法得到的最优参数不同,所构建的支持向量机分类模型都可实现正确的分类,且分类准确率均高达99.0%以上。研究结果表明,利用粒子群优化算法基于折射率构建的PCA-SVM分类模型效果最优,可以准确识别不同水样,为水质分类奠定了基础。     

基于小波系数图和卷积神经网络的太赫兹光谱物质识别

许多太赫兹光谱物质识别方法依靠寻找该物质在太赫兹波段范围内不同光谱表现出的不同特征来识别特定物质。吸收峰提取法是常用的光谱特征提取算法,但当光谱无明显特征吸收峰或峰位、峰值相近或难以识别时,难以利用吸收峰特征辨别物质。将机器学习和统计学习技术用于太赫兹光谱的识别中虽减少了吸收峰的干扰,但常常需要人为定义特征而导致分类误差。深度学习法能自动提取特征,但在识别前往往需要进行复杂的预处理操作,并且在特征提取的过程中容易丢失部分特征从而导致分类误差。针对以上问题,提出了一种基于小波系数图和卷积神经网络的太赫兹光谱识别方法。利用太赫兹光谱信号进行小波变换时,由于小波系数矩阵的每一行系数与原始光谱信号存在着对应关系,因此将太赫兹光谱的吸收系数通过小波变换在频率域上展开,能得到不同的二维的频率-尺度分布图,又称小波系数图。然后构造一个卷积神经网络(CNN)对小波系数图进行分类,可得到太赫兹光谱物质的分类结果。为了验证所提出算法的有效性,将三组小波系数图数据与原始光谱数据分别输入CNN、Support Vector Machin (SVM)、Multilayer Perceptron (MLP)三种不同的分类器作对比,从实验结果可以发现本文算法在三组数据中的识别率均达到了100%,说明相比于传统方法,本文方法能准确分类没有明显特征吸收峰的光谱,证明了使用卷积神经网络识别小波系数图的有效性。为了体现本文算法的优势,与小波脊线寻峰识别算法作对比,实验结果表明本文算法几乎不受峰频、峰位、峰值的影响,无论是识别不存在吸收峰的淀粉,还是识别相似度高的蔗糖和葡萄糖,都具有较高的识别率,分类准确率达97.62%,证明了所提算法的优越性。该算法为太赫兹光谱数据识别提供了一种新思路,同时也可以推广运用到其他谱图物质的识别中。     

密布式多输入多输出声呐阵列目标波达方向估计

针对低信噪比条件下多输入多输出声呐受对称噪声分量影响导致测向性能降低的情况,提出了一种基于协方差矩阵重构方法的波达方向估计算法。首先,将噪声场分为对称噪声和非对称噪声两部分,利用协方差矩阵虚部与对称信号无关的性质,去掉协方差矩阵的实部来降低对称噪声对目标波达方向估计精度的影响,采用降维转换方法和矩阵虚部置换原理重构协方差矩阵的实部,避免了双频谱的干扰。然后利用Toeplitz方法对重构的协方差矩阵进行解相干修正,通过奇异值分解获得噪声子空间,最后对目标的波达方向进行估计,可实现微弱信号的准确测向。理论分析和实验结果表明,该方法明显抑制了对称噪声,提高了目标的波达方向估计性能,具有运算速度快、自由度高和目标分辨力强的特点。