一种基于功率谱特征参量的水中目标辐射噪声非母板匹配分类识别方法

本文阐述了一种非母板匹配的水中目标辐射噪声分类识别的思想、技术路线及算法的计算机实现关键技术。该算法基于提取水中目标辐射噪声信号功率谱中的最强谱带位置、线谱位置、线谱根数、线谱平均间隔等特征参量，通过特定处理后，直接给出识别结果，不需要附带庞大数据库。经试验验证，该技术突破了传统的母板匹配目标识别的算法思想，能够适用于复杂海况，做到宽容性匹配。识别效率高，识别率达到试验预期要求。     

利用分频带空间谱和波束域的输出直流跳变与起伏比值融合检测未知线谱目标

目标辐射噪声中高强度稳定线谱可以用来提高常规宽带能量积分的检测性能。理论分析了已知线谱检测优于宽带能量积分检测的谱级比条件。给出了线谱频带可检测条件下,线谱谱级与最小可检测信噪比关系。分析分频带空间谱和波束域的输出直流跳变与起伏比值特征。在此基础上提出了一种构造频率、方位滤波矩阵的方法。该矩阵作用类似频带方位滤波器,通过对线谱频带目标方位的信号放大,与其他频带判决结果融合,来提高常规宽带能量积分的性能。最后给出该算法适用条件和性能分析。数据仿真和海试数据处理结果可知,该算法在多干扰源存在下对弱线谱目标提取能力强,且不需多帧统计,在检测效果上更接近于理想的窄带线谱检测方法。      

一种基于频率方差加权的线谱目标检测方法

在复杂的多目标声场中,常规波束形成（CBF）检测器性能显著下降。本文提出一种基于频率方差加权的线谱目标检测方法——CBF频率方差检测器,利用被动目标辐射噪声中含有高强度的稳定线谱这一特征,用每个方位频率域的峰值频率方差对CBF输出的方位谱进行加权。该检测器可在多目标强干扰和同波束强相干干扰背景中检测到弱线谱目标,且只须三维显示,避开了传统的线谱检测四维显示的难点。仿真和海试结果表明,在多目标、强干扰的环境下,可以探测到弱线谱目标。      

基于主副瓣比加权的未知线谱目标检测方法研究

针对目标辐射线谱信号未知时的目标检测问题,依据水下目标辐射噪声含有高强度稳定线谱这一特征,该文提出了一种基于波束形成主副瓣比加权的宽带波束形成目标检测方法。该方法首先用二阶锥优化各频带波束形成,得到低旁瓣高增益波束形成;其次利用各频率单元波束形成主副瓣差异形成加权因子;然后利用加权因子对各频率单元波束形成进行加权统计,可以抑制背景噪声能量干扰,增强目标检测信噪比增益,克服传统线谱检测四维显示难点。理论分析和实验结果表明：该方法可以较好地增强目标线谱单元能量、抑制噪声、提高信噪比,改善能量累积检测法在远程目标检测方面的性能。     

一种改进的时变转移概率AIMM跟踪算法

针对矢量线阵跟踪目标低频线谱提取问题,提出了一种矢量线阵低频线谱提取方法,其中利用拉平后线谱的均方差乘一比例因子设为门限的方法,可以有效提取线谱成分;并且研究了利用互谱中高信噪比线谱测向剔除干扰线谱方法,准确提取出目标特征线谱,可较有效解决矢量线阵左右舷目标低频线谱特征相互干扰问题。上述矢量线阵低频线谱提取方法,得到了海试数据的初步验证。     

被动线谱检测的子带分解和分方位区间融合算法

目标辐射噪声中低频线谱丰富,而且谱级比高、强度稳定,相比调制谱检测具有优越性。基于子带分解处理的现有融合方法适用在信噪比相对较高情况下,而当干扰是相干的或强宽带信号时,对线谱目标有效检测仍没很好解决。本文从子带空间谱统计特性出发,利用线谱谱级高出连续谱10-25dB;线谱频带所在方位区间的输出方位波动小,而其他分区间的输出方位波动大特点,提出了一种在信噪比低,多目标并存情况下更有效的弱线谱提取融合方法。理论仿真和海试实验数据处理结果验证该方法具有创新性、较常规方位稳定算法适用范围更广、检测效果更好,适合工程应用。     

舰船噪声识别(Ⅰ)──总体框架、线谱分析和提取

本系列文章的工作是在舰船噪声谱图的基础上，利用模糊神经网络对舰船进行分类识别。文章Ⅰ叙述了舰船噪声的谱特征由可分离的平稳谱和非平稳谱组成，介绍了既利用有效识别特征（有类别共性和异性的特征）又对特定舰船特征作专门记忆的工作路线及识别框架，对特定舰船的记忆具体体现在特定舰船特征模库一包含有线谱模板库、双重频率谱模板库和平均功率谱模板库。文章Ⅰ又具体讨论了特征提取和建立线谱模板时所碰到的理论、模型、分析参数及线谱提取方法，舰船噪声实际情况和理论之间的差异等问题。文章最后介绍了用机器自动提取线谱的一种方法。系列文章Ⅱ将讨论线谱稳定性、唯一性和线谱模板图；文章Ⅲ将讨论双重谱和平均功率谱的特征提取和模板建立；文章Ⅳ将讨论模糊神经网络和识别。     

遗传算法优化变分模态分解提取舰船辐射噪声特征线谱方法

特征线谱提取是舰船目标识别的一个重要研究环节,常采用传统的DEMON谱分析方法,处理过程中,一般对舰船噪声时域信号未予抑噪,低信噪比情况下,传统DEMON谱分析性能差。对此,提出一种采用遗传算法优化变分模态分解方法,用于分解舰船噪声原时域信号,获得抑制噪声后的舰船噪声重构信号,进而有效提取了舰船目标噪声幅度调制特征线谱。该方法首先采用遗传算法优化变分模态分解的两个关键输入参数(分解所取模态个数和惩罚因子),对变分模态分解得到的各阶固有模态分量加以判别,去除噪声主导分量,保留信号主导分量,使重构舰船噪声信号显著抑制了干扰噪声,然后对降噪后的重构信号进行频谱分析,获得目标噪声调制特征线谱。理论分析、仿真和实验数据处理结果表明,相比传统DEMON谱分析法,基于遗传算法优化变分模态分解的舰船噪声特征线谱提取方法具有更好的噪声抑制能力,所获取的舰船噪声幅度调制特征线谱信噪比明显高于传统DEMON方法,具有一定优势,前景良好。     

基于Karhunen-Loève变换和小波谱特征矢量量化的三维谱像数据压缩

提出了基于Karhunen Lo埁ｖｅ变换的小波谱特征矢量量化三维谱像数据压缩方法耍幔颍瑁酰睿澹?Lo埁ｖｅ变换 /小波变换 /小波谱特征矢量量化方法应用了Karhunen Lo埁ｖｅ变换的消除谱相关性优良性能 ,应用二维小波变换消除空间相关性 ,在小波变换域内应用二维集分割嵌入块编码和一维谱特征矢量量化对三维谱像数据压缩 ,获得较高的压缩性能。实验结果表明 :Karhunen Lo埁ｖｅ变换 /小波变换 /小波谱特征矢量量化编码比Karhunen Lo埁ｖｅ变换 /小波变换 /改进对块零树编码和Karhunen Lo埁ｖｅ变换 /小波变换 /快速矢量量化编码方法在同样压缩比条件下 ,峰值信噪比提高 2dB和 1dB以上 ,而速度提高了 1.5和 8倍 ,整体压缩性能有较大的提高     

层叠式“产生/判别”混合模型的语音情感识别

提出了层叠式“产生/判别”混合模型的语音情感识别方法。首先,提取63维语句级特征,运用Fisher从中选择12个最佳的语句级特征,建立小波神经网络(WNN)的层叠式产生式模型进行语音情感识别;然后提取69维帧级特征,采用SFS选择出待使用的8维特征,将高斯混合模型(GMM)进行多维概率输出,建立层叠式“产生/判别”混合模型进行语音情感识别。实验结果显示:(1)层叠式“产生/判别”混合模型较单独WNN、GMM、HMM (隐马尔可夫模型)、SVM (支持向量机)的识别率要高;(2)层叠式“产生/判决式”混合模型识别率较基于WNN的层叠产生式模型高;(3) M=13,D维GMM-MAP/SVM (MAP,最大后验概率)串联融合模型为最优的层叠式“产生/判别”混合模型,能获得最高85.1%的识别率。      

结合盲源分离的非直单线阵多目标左右舷分辨

针对单线阵左右舷模糊问题,提出了结合盲源分离的非直单线阵多目标左右舷分辨算法。该算法首先通过离散傅里叶变换将阵列接收信号离散化为若干个窄带频谱分量,之后对每个频点的窄带数据讲行盲源分离,得到每个频点上各个来波信号的导向向量;然后通过常规波束形成对各个导向向量进行方位谱估计,并根据左右舷抑制比进行单目标左右舷分辨,确定各个频点上的来波方向;最后对所有频点上的来波方向进行聚类,得到各个真实目标的方位,从而实现多目标左右舷分辨。仿真实验中,相比常规波束形成方法(CBF)和最小方差无畸变响应算法(MVDR),该算法更准确地估计出了目标数目,且保持了较快的计算速度;海试数据处理中,该算法排除了目标镜像的干扰,准确估计出了船只目标的轨迹。仿真及海试数据处理均表明,该算法可以分辨真实目标与目标镜像,具有比CBF和MVDR算法更好的左右舷分辨能力。      

Exo-atmospheric target discrimination using probabilistic neural network

Exo-atmospheric targets are especially difficult to distinguish using currently available techniques, because all target parts follow the same spatial trajectory. The feasibility of distinguishing multiple type components of exo-atmospheric targets is demonstrated by applying the probabilistic neural network. Differences in thermal behavior and time-varying signals of space-objects are analyzed during the selection of features used as inputs of the neural network. A novel multi-colorimetric technology is introduced to measure precisely the temporal evolutional characteristics of temperature and emissivity-area products. To test the effectiveness of the recognition algorithm, the results obtained from a set of synthetic multispectral data set are presented and discussed. These results indicate that the discrimination algorithm can obtain a remarkable success rate.     

阵不变量匹配主动声呐目标深度辨识

水面/水下目标深度辨识可以在主动声呐探测过程中快速筛选感兴趣目标。根据目标多径返回信号到达时间与目标深度的关联性,该文提出了基于水平线阵的主动脉冲响应匹配目标深度辨识方法,该方法利用主动信号传播特性结合阵不变量理论提取脉冲响应,并与拷贝声场匹配实现目标深度估计。在校正了目标多径回波的方位变化后,该方法提取的时域特征较为稳定,并且只在时延估计距离处进行深度搜索。通过水平阵收集的宽带实验数据,在连续的多次回波中实现了对水面/水下目标92%的辨识成功率,验证了该方法在浅海环境下的目标深度估计能力。     

基于时频图像分析的核爆与雷电电磁脉冲识别

根据核爆和雷电电磁脉冲信号非平稳、非线性特点,对核爆电磁脉冲(NEMP)和雷电电磁脉冲(LEMP)信号进行了Hilbert谱分析,计算了二者Hilbert谱的图像区域特征,对二者进行了识别研究,并且从NEMP和LEMP不同的产生机理上对识别结果进行了分析。实验结果表明:以Hilbert谱的面积和重心,以及六维图像区域特征作为特征,对NEMP和LEMP的识别率达到了90％以上,可以对二者进行有效识别。     

拖船噪声抵消与左右舷分辨联合处理方法的研究

利用拖曳双线阵声纳基阵结构的左右舷分辨能力,并结合拖船相对于拖曳线列阵的空间位置关系,本文提出一种拖船噪声抵消与左右舷分辨联合处理的方法,在波束域分别形成左右波束并将映像波束相减,在检测目标并进行左右分辨的同时,可以显著抵消拖船噪声。海试结果验证了该方法的有效性,拖船噪声被抑制掉约14dB,同时正确分辨出目标的左右舷方位。     

Lateralization of sinusoidally amplitude-modulated tones: effects of spectral locus and temporal variation

It has long been recognized that listeners are sensitive to interaural temporal disparities (ITDs) of low-frequency (i.e., below 1600 Hz) stimuli. Within the last three decades, it has often been demonstrated that listeners are also sensitive to ITDs within the envelope of high-frequency, complex stimuli. Because these studies, for the most part, employed discrimination tasks, few data exist concerning the extent of laterality produced by ITDs as a function of the spectral locus of the stimulus. To this end, we employed an acoustic "pointing" task in which listeners varied the interaural intensity difference of a 500-Hz narrow-band noise (the pointer) so that it matched the intracranial position of a second, experimenter-controlled stimulus (the target). Targets were sinusoidally amplitude-modulated tones centered on 500 Hz, 1, 2, 3, or 4 kHz and modulated at rates ranging from 50 to 800 Hz. Targets were presented with either the entire waveform delayed or with only the envelope delayed. Our results suggest that: (1) for low-frequency targets, lateralization is influenced by ITDs in the envelope but is dominated by ITDs in the fine structure; (2) for high-frequency targets, envelope-based delays produce displacements of the acoustic images which are affected greatly by the rate of modulation; rather large extents of laterality could be produced with high rates of modulation; these data are consistent with those obtained previously in discrimination experiments; (3) for low rates of modulation (e.g., 100 Hz), delays of the entire waveform (both envelope and fine structure) produce much greater displacements of the acoustic image for low-frequency than for high-frequency targets (where fine-structure-based cues are not utilizable); (4) there appear to be no consistent relations among extent of laterality, rate of modulation, and the frequency of the carrier within and across listeners.     

同色异谱破坏特性及其在相似异物检测中的应用

由于相似异物与所处背景的外观特征极其相近,现有的光谱成像检测技术遇到新的挑战。针对复杂背景中相似异物检测难题,提出了利用同色异谱破坏特性的透反射光谱图像检测方法。在分析同色异谱破坏过程中异物与背景光谱特性差异基础上,建立了波长优选函数筛选最优波长的透反射光源构建成像检测系统。以棉花中的白色丝状异物作为实验对象,利用最佳波长透反射成像系统获取异物特征,采用图像决策层融合增强异物图像特征,最后利用二值化处理提取异物目标。实验结果表明,在利用390 nm反射波长与580 nm透射波长的双波长图像检测中,成像系统获取的相似异物图像特征明显,异物提取后检测结果与实际相符。此方法可有效检测棉花中多种白色丝状异物,为相似异物检测提供了一种新思路。     

Foliage echoes: a probe into the ecological acoustics of bat echolocation

The research reported here aims at understanding the biosonar system of bats based on the properties of its natural inputs (ecological acoustics). Echoes from foliages are studied as examples of ubiquitous, natural targets. The echo properties and their qualitative relationship to plant architecture are described. The echoes were found to be profoundly stochastic and in general neither Gaussian nor stationary. Consequently, features useful for discrimination of such target classes will be confined to estimated random process parameters. Several such statistical signal features which are sufficiently invariant to allow a classification of the used example plants were identified: the characteristic exponent and the dispersion of an alpha-stable model for the amplitude distribution, a crest factor defined as the ratio of maximum squared amplitude and signal energy, the dispersion of the first threshold passage distribution, the structure of the correlation matrix, and a nonstationarity in sound channel gain. Discrimination error probability could be reduced by combining features pairwise. The best combination was the crest factor and the correlation coefficient of a log-linear model of the time-variant sound channel gain; it yielded an estimated Bayes risk of 6.9% for data pooled from different views.     

The functional analysis of auditory discrimination

Mammals have evolved the ability to acquire auditory discriminations. The characteristics of this discriminative ability presumably fit the natural conditions under which discriminations are normally acquired. The purpose of this paper is to review experiments which were directed at showing that auditory discriminations are most rapidly acquired when natural features are incorporated into the experiments. The experiments were also directed at discovering the underlying characteristics of the discriminative ability. When animals were trained to discriminate the position of a sound source in which natural features were incorporated into the experiment, the discrimination was acquired in one trial. Manipulation of the natural features suggested that one trial acquisition depends upon the following. (1) Stimulus novelty; the effect of reinforcement is stronger in the presence of novel than familiar stimuli. (2) Specific behavioral effect of reinforcement; the effect of reinforcing a response in the presence of a novel auditory stimulus is to increase the strength of approaching and manipulating the sound source.     

Insights into dolphin sonar discrimination capabilities from human listening experiments

A variety of dolphin sonar discrimination experiments have been conducted, yet little is known about the cues utilized by dolphins in making fine target discriminations. In order to gain insights on cues available to echolocating dolphins, sonar discrimination experiments were conducted with human subjects using the same targets employed in dolphin experiments. When digital recordings of echoes from targets ensonified with a dolphinlike signal were played back at a slower rate to human subjects, they could also make fine target discriminations under controlled laboratory conditions about as well as dolphins under less controlled conditions. Subjects reported that time-separation-pitch and duration cues were important. They also reported that low-amplitude echo components 32 dB below the maximum echo component were usable. The signal-to-noise ratio had to be greater than 10 dB above the detection threshold for simple discrimination and 30 dB for difficult discrimination. Except for two cases in which spectral cues in the form of "click pitch" were important, subjects indicated that time-domain rather than frequency-domain processing seemed to be more relevant in analyzing the echoes.