贝叶斯优化卷积神经网络公共场所异常声识别*

针对公共场所异常声的感知和识别问题,提出一种基于贝叶斯优化卷积神经网络的识别方法。提取声信号的Gammatone倒谱系数、倍频程功率谱、短时能量和谱质心,组合成声信号的特征图。构建卷积神经网络作为分类器,利用递增的卷积核设置和池化操作处理不同尺度的特征。基于贝叶斯优化算法优化卷积神经网络的模型参数,对包括火苗噼啪声、婴儿啼哭声、烟花燃放声、玻璃破碎声和警报声的5种公共场所异常声进行识别。该方法的识别结果与基于不同的特征提取和分类器方案得到的识别结果进行比较,结果表明该方法的识别效果优于其他特征提取和分类器方案的识别效果。最后分析了该方法在不同信噪比噪声干扰下的识别结果,验证了该方法的有效性。     

用多消失矩最优小波包基改进虹膜识别

采用层叠算法求出尺度函数和小波函数的离散采样序列的逼进，将不具有基函数解析表达式的母小波用于虹膜特征提取，并利于实现连续小波包变换.针对虹膜光学识别应用背景，提出采用基于连续小波包变换的多消失矩联合最优小波包基来改进特征图像相关识别的最优基，并用统计识别方法进行后处理以增强算法适应性，同时提出通过体全息相关系统来实现，以发挥光学高并行性的优势，模拟结果表明可获得比已有方法更高的识别率.     

Hybrid multi-resolution detection of moving targets in infrared imagery

A hybrid moving target detection approach in multi-resolution framework for thermal infrared imagery is presented. Background subtraction and optical flow methods are widely used to detect moving targets. However, each method has some pros and cons which limits the performance. Conventional background subtraction is affected by dynamic noise and partial extraction of targets. Fast independent component analysis based background subtraction is efficient for target detection in infrared image sequences; however the noise increases for small targets. Well known motion detection method is optical flow. Still the method produces partial detection for low textured images and also computationally expensive due to gradient calculation for each pixel location. The synergistic approach of conventional background subtraction, fast independent component analysis and optical flow methods at different resolutions provide promising detection of targets with reduced time complexity. The dynamic background noise is compensated by the background update. The methodology is validated with benchmark infrared image datasets as well as experimentally generated infrared image sequences of moving targets in the field under various conditions of varying illumination, ambience temperature and the distance of the target from the sensor location. The significant value of F-measure validates the efficiency of the proposed methodology with high confidence of detection and low false alarms.     

Structural partitioning of complex structures in the medium-frequency range. An application to an automotive vehicle

In a recent work [Journal of Sound and Vibration 323 (2009) 849-863] the authors presented an energy-density field approach for the vibroacoustic analysis of complex structures in the low and medium frequency ranges. In this approach, a local vibroacoustic energy model as well as a simplification of this model were constructed. In this paper, firstly an extension of the previous theory is performed in order to include the case of general input forces and secondly, a structural partitioning methodology is presented along with a set of tools used for the construction of a partitioning. Finally, an application is presented for an automotive vehicle.     

Solitary electromagnetic pulses in a retarding system with an electron-ion beam

The propagation of a combined electron-ion beam in a retarding plasma-filled system under the conditions of cyclotron (abnormal Doppler) resonance of electrons with the wave is studied. We show the possibility of existence of resonant electromagnetic solitons, which propagate along the beam. The nonlinear dispersion law and group velocity of the soliton are determined. Special attention is paid to the case in which effects associated with the longitudinal polarization field, which appears during charge separation in the electromagnetic-pulse field, are important. By numerical integration we have found solutions of the solitary-wave type, which exist for the discrete spectrum of system parameters. For the limiting case of rarefied plasma, when the discrete spectrum becomes continuous, the solution is found in the adiabatic approximation.Translated from Izvestiya Vysshikh Uchebnykh Zavedenii, Radiofizika, Vol. 39, No. 7, pp. 850–866, July, 1996.     

A One-Dimensional Discrete Boltzmann Model for Detonation and an Abnormal Detonation Phenomenon

A one-dimensional discrete Boltzmann model for detonation simulation is presented. Instead of numerical solving Navier-Stokes equations, this model obtains the information of flow field through numerical solving specially discretized Boltzmann equation. Several classical benchmarks including Sod shock wave tube, Colella explosion problem,and one-dimensional self-sustainable stable detonation are simulated to validate the new model. Based on the new model,the influence of negative temperature coefficient of reaction rate on detonation is further investigated. It is found that an abnormal detonation with two wave heads periodically appears under negative temperature coefficient condition.The causes of the abnormal detonation are analyzed. One typical cycle of the periodic abnormal detonation and its development process are discussed.     

云计算下石油化工罐区储罐液仪表设计与实现

近年来,随着我国互联网云计算的不断发展,云计算技术被应用于诸多领域。针对传统石油化工领域中储罐液仪表存在的罐内压力检测精准度差、温度感应灵敏度低与储罐液状态数据分析面窄等问题,提出云计算下石油化工罐区储罐液仪表设计。通过大数据运算核心进行框架构建,在此基础上,采用PCJD气流密度差算法、微感热源运算单元与数据流云算技术,对传统储罐液仪表进行设计改进,从问题的根源进行针对性解决。仿真对比试验证明,提出的云计算下石油化工罐区储罐液仪表设计,具有罐内液压力数据反馈精准度高,温度感应灵敏度高、储罐液综合数据分析速度快、数据分析完整度高等优点。     

Discrete and selective absorption in crystalline molecular nanofilms

Recent research in nano-optical engineering and in nanomedicine as well, seeks for methods of construction of various types of nano-markers, nano-carriers, and ways to deliver drugs to the exactly determined regions of body. In this process it is important to find methods of recognition of certain types of molecules. It is obvious that optical recognition would be the easiest and the most effective way to do it. Our research presents a model of a molecular ultrathin crystalline film and generated exciton system inside it and corresponding methodology of analysis of their optical characteristics. Properties of these spatially very restricted structures are very sensitive to their surrounding surfaces. Using the two-time Green’s functions adapted for crystalline structures with symmetry breaking, and graphical-numerical software, we have calculated the energy spectra and possible exciton states. We have shown that the appearance and the presence of localized states on the surfaces and in the boundary layers of the film depend on the thickness of the film and the film surroundings, presented through the perturbation of parameters on surfaces. Optical properties in these structures demonstrate discrete and very selective resonant absorption spectra, depending on the perturbation on their surfaces.     

窗口可变滑动相关分析方法在激光诱导击穿光谱谱线自动识别中的应用

激光诱导击穿光谱(LIBS)谱线具有很窄的展宽,且光谱仪因外界温度条件的变化存在波长漂移,光谱数据为离散数字信号,受噪声、谱线重叠及连续背景的影响,信号存在失真和变形。而目前光谱仪自带谱线识别分析软件主要以"就近原则"为主,错误率较高,必须依赖于人眼的观察对比。针对这个问题,在研究LIBS光谱特性的基础上,提出了一种窗口可变滑动相关分析方法(CAASW),用于激光诱导击穿光谱谱线自动识别。以土壤标准物质为样品,对该方法进行了实验分析和评价,与光谱仪自带分析识别软件相比,CAASW明显提高了识别准确率和识别速度,最终实现了谱线的自动识别。     

Discrete channel modelling based on genetic algorithm and simulated annealing for training hidden Markov model

Hidden Markov models (HMMs) have been used to model burst error sources of wireless channels. This paper proposes a hybrid method of using genetic algorithm (GA) and simulated annealing (SA) to train HMM for discrete channel modelling. The proposed method is compared with pure GA, and experimental results show that the HMMs trained by the hybrid method can better describe the error sequences due to SA's ability of facilitating hill-climbing at the later stage of the search. The burst error statistics of the HMMs trained by the proposed method and the corresponding error sequences are also presented to validate the proposed method.     

基于假设检验的小弱运动目标航迹起始

根据目标与噪音(或虚警)在时域和空域上所表现出的不同特性,提出了一种基于空域邻域聚类和时域轨迹连续性相结合的假设检验方法,以建立小弱运动目标航迹起始,并对其性能进行了详细分析.基本出发点是:空域上,小目标是占有少量像素且幅度均匀的小块,聚集度较高,而噪音点是独立、不相关的,聚集度较低;时域上,运动目标将以较大概率出现在由轨迹连续性决定的预测点邻域内,而虚警是杂乱无章,连续多次出现在该邻域内的几率较小.理论分析和实验结果表明,本文方法能较为有效地建立目标初始航迹.     

SUNISTװ�õ�ECW����ŷķ��Ӧ�������ʵ��

介绍了在SUNIST装置上进行的ECW辅助欧姆感应电流启动实验。在ECW(2.45GHz/20kW/10ms)和垂直场感应的共同作用下,先驱动起一个较小的等离子体电流(<1kA),再投入欧姆场,可以获得比常规欧姆放电更高的电流爬升速率,提高伏秒数的利用效率。这种启动方式为SUNIST欧姆放电提高了灵活性,可以在不进行硬件改动的前提下增加调节等离子体平衡的余量,拉长放电的持续时间。     

Noether's theory of Lagrange systems in discrete case

In this paper, Noether theory of Lagrange systems in discrete case are studied. First, we briefly overview the well-known Noether theory of Lagrange system in the continuous case. Then, we introduce some definitions and notations, such as the operators of discrete translation to the right and the left and the operators of discrete differentiation to the right and the left, and give the conditions for the invariance of the difference functional on the uniform lattice and the non-uniform one, respectively. We also deduce the discrete analog of the Noether-type identity. Finally, the discrete analog of Noether's theorem is presented. An example was discussed to illustrate these results.     

物联网环境下的差异网络数据库异常数据检测

当前的物联网环境下,各个智能网络的数据库的使用没有统一标准,不同生产商的数据库中的异常数据标准也不同,这就使得传统的以模式识别为基础的网络数据库异常检测方法在进行异常阀值设置时,无法形成统一标准,数据库数据量庞大且存在无序性,无法保证检测的准确性和检测效率。提出基于混沌特征分析算法的物联网环境下的差异网络数据库异常数据检测方法。依据混沌特征分析相关理论构建物联网环境下的差异网络数据库模型,构建一种异常数据的偏差函数,对不同数据库下的异常数据进行偏差统计,通过对偏差函数的统计结果进行最小值求解,根据求解描述最小化的阀值请求,实现物联网环境下的差异网络数据库异常数据的检测。实验结果表明,利用改进算法进行异常数据检测,能够提高检测的有效性与准确性。     

Vertex Operator Solutions to the Discrete KP-Hierarchy

Vertex operators, which are disguised Darboux maps, transform solutions of the KP equation into new ones. In this paper, we show that the bi-infinite sequence obtained by Darboux transforming an arbitrary KP solution recursively forward and backwards, yields a solution to the discrete KP-hierarchy. The latter is a KP hierarchy where the continuous space $x$-variable gets replaced by a discrete $n$-variable. The fact that these sequences satisfy the discrete KP hierarchy is tantamount to certain bilinear relations connecting the consecutive KP solutions in the sequence. At the Grassmannian level, these relations are equivalent to a very simple fact, which is the nesting of the associated infinite-dimensional planes (flag). The discrete KP hierarchy can thus be viewed as a container for an entire ensemble of vertex or Darboux generated KP solutions. It turns out that many new and old systems lead to such discrete (semi-infinite) solutions, like sequences of soliton solutions, with more and more solitons, sequences of Calogero–Moser systems, having more and more particles, just to mention a few examples; this is developed in [3]. In this paper, as another example, we show that the q-KP hierarchy maps, via a kind of Fourier transform, into the discrete KP hierarchy, enabling us to write down a very large class of solutions to the q-KP hierarchy. This was also reported in a brief note [4]. Received: 27 August 1998 / Accepted: 24 November 1998     

Discrete variational principle and first integrals for Lagrange--Maxwell mechanico-electrical systems

This paper presents a discrete variational principle and a method to build first-integrals for finite dimensional Lagrange--Maxwell mechanico-electrical systems with nonconservative forces and a dissipation function. The discrete variational principle and the corresponding Euler--Lagrange equations are derived from a discrete action associated to these systems. The first-integrals are obtained by introducing the infinitesimal transformation with respect to the generalized coordinates and electric quantities of the systems. This work also extends discrete Noether symmetries to mechanico-electrical dynamical systems. A practical example is presented to illustrate the results.     

Exact 1+1 dimensional solutions of the discrete velocity Broadwell-Boltzmann model

For the 6-velocity Broadwell model with four independent densities, exact 1+1 dimensional solutions (space x, time t) are found. They are sums of two similarity waves with components either real or complex conjugate. The first case provides 1+1 dimensional shock-waves. In the second case we build periodic solutions in the space which either are propagating or nonpropagating in time. The same method has been successfully applied to other discrete models.     

Augmented Lanczos bidiagonalization by small singular value decompositions for face recognition and image compression

A new method for face recognition and image compression is presented here. In this method, Lanczos bidiagonalization is combined with sequences of small singular value decompositions. The goal of this method is to perform the truncated singular value decomposition for large matrices. It is an extension of standard Lanczos bidiagonalization, which improves its approximation capabilities and keeps the computational cost reasonable. Face recognition and image compression examples show the new method is competitive with pervious methods.