增强型双通道实时联想存储系统的研究

首次将多通道并行处理的概念引入联想存储系统中，提出一个带有ＣＣＤ摄像机、图像卡、微机及液晶光阈等控制与输入设备，并以液晶电光开关作为取阈反馈装置的双通道联想系统。系统存储方式采用计算机磁盘充当外部图像库存储大量参考图像，而在系统内一点只存储一幅图像。     

局域互联神经网络的关联存储

基于全局互联的Hopfield模型,本文提出了局域互联关联存储的新概念.与全局互联相比,局域互联具有较小的关联矩阵,因而,有利于用现有的空间光调制器加以实现.同时,计算机模拟结果表明,它仍然具有全局关联存储的能力.     

多值神经网络改进模型及其光学实现

本文提出一种改进的光学神经网络模型,并利用空间光调制器PROM构成的光学系统实现了这种模型的联想记忆运算.计算机模拟和实验结果表明,改进模型提高了光学神经网络的识别能力,并在—定程度上提高了存贮容量.     

两维局域互联神经网络的关联存储

本文将局域互联神经网络的新概念推广到两维情形,并对两维局域互联关联存储进行了理论分析和大量的计算机模拟.结果表明,两维局域互联神经网络的优点是,在满足存储容量限制的前提下,它与全局互联神经网络具有相同的关联存储能力,而其互联权重矩阵要比全局互联网络小得多.因而,有利于使用现有的空间光调制器实现两维大规模的人工神经网络.     

双极双向联想存储器的单通道光学实现

以附加神经元引入附加背景的方式获得将线性离散比极神经元的神经网络在单通道光学矢量-矩阵乘法器内实现的方法,给出了相应的光学系统的修正和非负光学模板的编码形式.以双极神经元的双向联想存储器为例进行了计算机和光电实验模拟.     

一维局域互联关联存贮的光学实现

本文提出了一维局域互联关联存贮的光学实现方法，讨论了可用来实现局域互联网的三种光电混合系统。基于光电相关系统并用图象监示器编码互联权重矩阵，液晶显示器编码输入矢量，给出了一维光学实验结果。     

采用自泵浦相位共轭镜的光学关联存储器

本文报道了一种采用自泵浦相位共轭镜作为取阈,反馈器件的光学关联存储系统,该系统采用He-Ne激光作为光源,可实现实时处理.当输入信息仅为存储信息的25%时,得到了正确的输出.     

多状态,多阈值神经网络模型的光电混合实现

提出了一种采用高分辨率液晶电视（ＬＣＴＶ）实现Ｈｏｐｆｉｅｌｄ神经网络多值算法的光电系统，文章给出了平面多状态，多阈值的全互连Ｈｏｐｆｉｅｄ神经网络模型，并采用该系统对颜色进行了联想和记忆的实验，初步的实验结果可以证实，此种高分辨率液晶电视神经网络系统是可行的。     

Utilizing nonlinear optical pattern formation for a simple image-processing task

A novel scheme for all-optical image processing is suggested, which is based on spontaneous pattern-formation processes. The spatiotemporal instabilities, connected with these self-organization effects, appear in many nonlinear optical systems and mostly obstruct the intended application. We propose to utilize the particular features of pattern formation for parallel image processing, namely the sensitivity to external influences, the selection of well-defined final states, and the rotational and translational invariance. We report here on a single-feedback experiment with a liquid crystal light valve as the optical nonlinearity. In this experiment, the recognition of hexagonal structures is realized. We characterize the response dynamics and study associative properties of this scheme. The extension to square patterns is discussed, and first steps towards a practical implementation are undertaken in providing a simple post-processing scheme and testing the concept on realistic input images. Received: 26 November 2002 / Revised version: 7 February 2003 / Published online: 9 April 2003 RID="*" ID="*"Corresponding author. Fax: +49-6151/16-4123, E-mail: ralph.neubecker@physik.tu-darmstadt.de     

SET-MOS混合结构的细胞神经网络及其应用

基于细胞神经网络（CNN）细胞单元的等效电路及其电学特性模型,利用SET-MOS混合结构反相器实现了模型中的激活函数电路,用耦合电容单元实现CNN细胞的系统模板,构建了SET-MOS CNN细胞硬件电路,并将其应用在图像处理中.仿真结果表明,所设计的CNN硬件电路具有结构简单、功耗低、响应速度快等特点,可用于构成各种规模的CNN电路,进一步满足大规模信号处理的需求及提高集成电路的集成度. 关键词： 单电子晶体管 MOS管 细胞神经网络 图像处理     

Errors of neural networks

Errors of associative sampling and recognition in neural networks are determined by methods of the general theory of statistical solutions taking into account statistical variation of the input image, the correlation radius of the image, the number of neurons, and the number of images in the memory. The beta-distribution is proposed to be used as the statistics of image variations. The admissibility of the utilization of the distribution is tested by using temporal variations of real scenes as an example. The error in determining the coordinate is found for networks on the basis of optical correlators.     

Overview of hybrid optical neural networks

This paper reviews optical architectures for the implementation of hybrid neural networks. Optics is mainly applied to implementing the matrix-vector or tensor-matrix multiplication. In addition, the general background of neural networks as well as a brief discussion on holographic associative memory are also given.     

Optical fuzzy systems based on efficient implementation of a fuzzy associative memory

We propose efficient implementation of a fuzzy associative memory system based on a fuzzy-sets decomposition technique. The technique involves decomposing the input fuzzy sets of a fuzzy associative memory into orthogonal subsets so multiple fuzzy rules can be superimposed without loss of information. This technique can significantly reduce the optical hardware and computation needs of fuzzy associative memory systems.     

基于混合类间联想神经网络的光学模式识别

通过结合自联合模型与异联想模型，提出了一种混合类间联想神经网络的光学模式识别系统。在该神经网络中，输入模式矢量不仅通过自联想识别自身，还通过异联想识别配对矢量，因而提高了识别概率。与匹配滤波器光学模式识别系统相比，识别结果直接，系统简单可行。     

Simulated quantum-optical object recognition from high-resolution images

A holographic experimental procedure assuming use of quantum states of light is simulated. It uses merely interference-based image storage and nonunitary image retrieval realized by wave function collapse. Successful results of computational view-invariant recognition of object images are presented. As in neural net theory, recognition is selective reconstruction of an image from a database of many concrete images (simultaneously stored in an associative memory) after presentation of a different version of that image. That is, in the first step, we store many high-resolution images of objects into quantum memory (a hologram). In the second step, we present a “nonlearned” noisy image version. We thereby trigger memory-influenced reorganization of the state of the system so that it finally encodes those corrected object images that correspond to the newly presented version. The holographic procedure seems to be implementable with present-day quantum optics.     

Stimulated photon echo with temporal data coding in the Hopfield-Little regime

A model of an optical neural network with associative memory using stimulated photon echo (SPE) with temporal data coding is presented. The electrical field’s calculation of the SPE signals is carried out for case of multiple influences on the resonant medium by optical pulses. A computer model of an optical echo processor with associative retrieval of information is elaborated.     

Sequential associative information storage and reconstruction in a holographic circuit

After a brief discussion of the analogy between neutral and holographic information processing, a holographic circuit for associative information storage and reconstruction is described. This special system models the ability of the brain to reconstruct a sequence of previously stored informations A, B, C,…, when only one input stimulus is presented to the neural system. Furthermore the holographic system is capable of retrieving branching information, modelling the property of the brain to memorize various sequences of information with one special input.     

三种光电混合神经网络系统衍射误差的仿真研究

介绍了三种类型的光电混合神经网络系统,重点研究了衍射引起的误差对系统输出的影响。根据衍射、光学信息处理和神经网络理论,采用实验中的参量和输入数据,对衍射造成的输出误差做了仿真分析。分析表明,近场衍射造成较大的输出误差。近场与远场衍射综合作用时,误差因输入图像模式不同而差异较大;其中,当输入较小的简单图像时相对误差较大;而对实验中实际采用的复杂图像,相对误差较小。利用线性回归方法对输出数据做了校正,并分析了其可行性,校正后的数据误差降低一个量级。衍射误差对实验中识别率的影响可以控制在较小的范围内,识别率可以保持在97.7%以上。     

Optical Associative Memory Based on Double Phase Conjugation Using Bi_（12）TiO_（20） Photorefractive Fi

ＯｐｔｉｃａｌＡｓｓｏｃｉａｔｉｖｅＭｅｍｏｒｙＢａｓｅｄｏｎＤｏｕｂｌｅＰｈａｓｅＣｏｎｊｕｇａｔｉｏｎＵｓｉｎｇＢｉ＿（１２）ＴｉＯ＿（２０）ＰｈｏｔｏｒｅｆｒａｃｔｉｖｅＦｉｂｅｒ￥ＬＩＵＳｈｕｔｉａｎ（ＤｅｐａｒｔｍｅｎｔｏｆＰｈｙｓｉｃｓ，...     

光计算中关联存储器的一种新模式

本文提出了光计算中关联存储器的一种新的存储模式,称为自反关联存储器。在这种模式中,矢量是以矢量对的形式被贮存在存储矩阵中。用预先贮存的矢量对中的一个矢量的部分信息不但能取出这个矢量的完整矢量,而且能取出这个矢量对中的另一个矢量。文中并提出了自反关联存储器的光学实现方法。