ARTMAP与模糊逻辑结合的字符智能识别方法

研究了将ＡＲＴＭＡＰ神经网络与模糊规则相结合的字符识别方法．ＡＲＴＭＡＰ用于对字符的学习、训练，而模糊规则和ＡＲＴＭＡＰ共同用于对字符的智能识别．实验证明这种将模糊技术与神经网络相结合的混合系统具有较高的识别率和较快的识别速度，采用ＡＲＴＭＡＰ神经网络只需要较少的网络训练时间．     

基于混合神经网络的车牌字符识别技术

车牌识别是智能交通系统的重要组成部分,其关键是车牌字符识别技术。单一的神经网络难以识别模糊的车牌字符,文中提出了一种混合神经网络实现车牌字符识别技术。该混合神经网络结合联想记忆与BP神经网络,对输入的字符进行两次判别,经过训练、特征提取得到检测结果。通过在不同的噪声和不同的角度实验表明,采用混合神经网络具有更高的识别精度。     

基于RBF的纸币序列号识别方法

序列号识别是金融安全领域的一个重要步骤.针对纸币序列号的字符特点,提出了一种基于径向基函数神经网络模型(RBF)的光学字符识别方法.使用工业相机,采集得到纸币图像,并进行必要的预处理,得到序列号区域的二值图像.再根据连通区域信息对单个字符进行分割并归一化,然后提取字符的二维空间分布信息作为特征向量,并结合径向基函数神经网络进行字符识别.实验证明,该神经网络仅两层隐藏层,结构简单,在小样本集下可达到95%的识别率,并且训练时间在100ms量级,能够很好地满足纸币序列号识别功能的要求.     

用于字符识别的椭圆基函数概率神经网络

提出了一种适合在有噪声条件下进行字符识别的椭圆基函数概率神经网络ＥＢＰＮＮ模型，该模型选用由字符统计特征提取的具有选择注意特性的参数作为概率神经网络的椭圆基函娄系数。ＥＢＰＮＮ模型在货车编号自动识别系统中获得了良好的应用，整个系统的字符识别率达到９６％以上，编号识别率达到９０％以上，实验结果表明该模型的识别性能较常用的其它神经网络要好，且特别适用于有噪声的情况。     

基于BP神经网络的文本验证码破解

以某游戏网站的验证码为例,介绍了基于文本图像验证码的识别和破解过程。破解过程分为字符提取、字符修正和字符识别。在字符提取阶段需要对图片进行预处理降低提取难度,提取主要采用了近似颜色统计法;在字符修正阶段对比了传统的旋转算法和改进旋转算法,然后归一化字符;在字符识别阶段采用了BP神经网络方法,对验证码的识别正确率达70％,每个验证码的平均破解时间为1．625s。     

人工免疫系统在手写希腊字符识别中的应用

基于自然免疫系统的人工免疫系统是人工神经网络和进化计算之后的新型计算智能,以其强大的信息处理和问题解决能力被广泛应用到众多领域。本文提出了基于人工免疫系统的手写希腊字符识别策略并进行了实验仿真。利用3*3子网格算法对24*24字符图像样本提取64个特征参数,通过人工免疫算法训练图像样本,建立24个抗体库用于识别手写希腊字符。另外,利用BP神经网络进行了对比实验。实验结果表明,人工免疫系统模型在手写希腊字符识别中较BP神经网络具有更大的优势。     

基于神经网络集成的车牌字符识别

车牌字符识别技术是图像识别领域的一个重要研究课题,在完成对车牌图像预处理、定位和分割的基础上,提出一种基于多个神经网络分类器综合集成的车牌字符识别方法.将基于粗网格特征提取的BP分类器、基于局部灰度特征提取的RBF分类器和基于方向线素特征提取的SOFM分类器通过串并联的方式进行集成,用Matlab7.0进行实验,通过实验证明多神经网络集成方法的可行性,该方法有效提高了整个系统的抗干扰性和字符的识别率.     

基于小波特征的单字符汉字字体识别

汉字图像不仅包含了汉字的字符信息,还包含了汉字的字体信息.字体信息是版面分析、理解和恢复的重要依据,还有助于实现高性能字符识别系统.目前的字体识别方法还不能对单个汉字字符的字体进行识别.本文提出了一种新的字体识别方法,能够在不知道汉字字符的前提下,识别单个汉字的字体.首先对单个汉字的字符图像进行小波分解,在变换图像上提取小波特征.提取的小波特征经Box-Cox变换整形后,用线性鉴别分析技术(LDA)进行特征选择,得到字体识别特征.所使用的分类器是MQDF分类器.在包含7种字体的样本集上进行的实验表明,本文提出的方法能够在不知道汉字字符的前提下,对单个汉字的字体进行有效识别,基于单字的字体识别率达到97.35％.     

一种车牌识别软件系统研究

针对现有车牌识别算法中的车牌定位、字符分割和字符识别三个核心模块存在的不足,提出了一种基于边缘两侧颜色检测的车牌定位方法;通过采用动态阈值调整方法,很好的实现了字符分割;对神经网络在字符识别技术中的应用进行了大量的研究和实验,根据汉字和数字、字母特征提取的不同,在对字符信息初识别时将汉字和数字、字母采用不同结构参数设置的神经网络进行识别,并对识别结果中包含的具有形体相近的字符提出了一种"不等权值"的方法。结果证明识别率有了明显提高。     

基于EMD的车牌字符去噪校准算法

对典型的车辆牌照字符识别算法中的去噪校准算法进行了研究和分析,在此基础上提出一种新型去噪算法,通过提取分割字符的积分投影波形进行EMD与SG联合去噪后重构字符,克服了字符特征模糊的缺陷,为车牌字符识别提供了可靠的字符特征。     

Membership function learning in fuzzy classification

A novel three-layer fuzzy neural network (FNN) is proposed which possesses ihe structure and learning ability of artificial neural networks, and the classification ability of fuzzy algorithms for pattern classification problems. During learning, the proposed FNN learns the membership function of each fuzzy class from training samples and adaptively organizes its hidden layer. The learning and recall times of the FNN are fast. Simulation results are also presented.     

Modified back-propagation algorithm applied to decision-feedback equalisation

This study presents a modified back-propagation (BP) algorithm for a multilayer perceptron (MLP) to perfect its ability to cope with the problem of binary phase shift keying channel equalisation. For a typical BP algorithm, the error signal is obtained from the comparison between the target and estimated signal. The error signal is propagated layer by layer from the output layer to the input layer to adaptively adjust all weights in the MLP. Therefore all parameters of the MLP are obtained by a single BP algorithm. However, the structure of the MLP with a hidden layer provides the feasibility to modify the BP algorithm to improve its performance. The MLP can be divided from the hidden layer into two sub-MLPs, and each sub-MLP is optimised by its own BP algorithm. Accordingly, the whole MLP is adjusted by two BP algorithms independently. In this study, the modified BP algorithm is utilised to cope with the problem of channel equalisation. The simulation results show that the modified BP algorithm indeed improves the typical BP algorithm especially for an environment with nonlinear distortion, frequency offset, and phase and timing errors. Moreover, the computation complexity of the proposed algorithm almost equals that of the conventional BP algorithm     

HBP: improvement in BP algorithm for an adaptive MLP decision feedback equalizer

Though the decision feedback equalizer (DFE) with multilayer perceptron (MLP) structure can be trained effectively by the backpropagation (BP) algorithm, it is always accompanied by the problem of local minimum. In order to solve some problems of the local minimum in the BP algorithm and to improve the performance of the BP algorithm under the same MLP structure, we combine the hierarchical approach and the BP algorithm to implement the MLP DFE, and we call the new scheme hierarchical BP (HBP) algorithm. Based on the hierarchical approach, from the input layer to the output layer of the MLP, every two layers of neural nodes (with one hidden layer) will be trained with an individual BP algorithm. Therefore, the entire MLP can be trained by several independent BP algorithms, unlike the standard BP algorithm, which utilizes only one BP algorithm to train the whole MLP structure. The results of performance evaluation indicate that the HBP algorithm not only strongly reduces the mean squared error but also yields a much lower bit-error rate than the standard BP algorithm does for equal computational cost and conditions.     

多媒体通信中智能化媒体内同步机制

本文提出了一种智能化视频流量的预测和同步机制（ＩＦＳＭ）,它由ＢＰ神经网络流量预测器（ＢＰＮＮ）、输出缓冲区和基于模糊神经网络（ＦＮＮ）的输出速率决策器所组成。ＢＰＮＮ采用一种在线训练的ＢＰ神经网络预测在将来的一定时间间隔（ＦＩ）内的平均分组速率,ＦＮＮ决策器根据预测的流量特性和缓冲区中的分组数动态地调节下一个分组输出的时间。仿真结果表明：与窗口机制相比,ＩＦＳＭ能够使视频流量取得较高的连续性和较低的时延,并且由于ＦＮＮ的学习能力,ＩＦＳＭ可以自适应地调节相应参数以满足不同的服务质量的要求。     

The Hidden Markov Model of co-articulation and its application to the continuous speech recognition

The co-articulation is one of the main reasons that makes the speech recognition difficult. However, the traditional Hidden Markov Models(HMM) can not model the co-articulation, because they depend on the first-order assumption. In this paper, for modeling the co-articulation, a more perfect HMM than traditional first order HMM is proposed on the basis of the authors' previous works(1997, 1998) and they give a method in that this HMM is used in continuous speech recognition by means of multilayer perceptrons(MLP), i.e. the hybrid HMM/MLP method with triple MLP structure. The experimental result shows that this new hybrid HMM/MLP method decreases error rate in comparison with authors' previous works.     

故障诊断交流字典法的前向神经网络实现方法

提出了一种采用ＢＰ算法的前向多层神经网络实现模拟电路故障诊断交流字典的方法。     

Intelligent Video Traffic Smooth Mechanism for Multimedia Communication

1 Introduction Broad-band integrated service digital networks (B-ISDN) based on the asynchronous transfer mode (ATM) are designed to support a wide variety of multimedia with diverse statistical characteristics and quality of services (QoS). Among the var…     

Intelligent Video Traffic Smooth Mechanism for Multimedia Communication

A principal challenge in supporting real-time video services over ATM is the need to provide synchronous play-out in the face of stochastic end-to-end network delays. In this paper, an intelligent traffic smooth mechanism ( ITSM ) is proposed to meet the continuity requirement which is composed of a back-propagation neural network ( BPNN ) traffic predictor, a play-out buffer, and a fuzzy neural network ( FNN ) based play-out rate determinator. The BPNN traffic predictor online predicts the mean packet rate of the traffic in the future interval ( FI ) and the FNN is designed to adaptively determinate the play-out time according to the number of packets in the buffer and the traffic character predicted. Simulation results show that compared to the window mechanism, ITSM achieves high continuity with accepted delay. Furthermore, ITSM can be adaptively modified to meet the QoS of different kinds of services by FNN parameter training.     

Peak‐to‐average power ratio reduction in space–time coded MIMO‐OFDM via preprocessing

In this paper, we propose a trellis exploration algorithm based preprocessing strategy to lower the peak‐to‐average power ratio (PAPR) of precoded MIMO‐OFDM. We first illustrate the degradation in PAPR due to optimal linear precoding in MIMO‐OFDM systems. Then we propose two forms of multi‐layer precoding (MLP) schemes to reduce PAPR. In both schemes, the inner‐layer precoder is designed to optimize system capacity/BER performance. In the first MLP scheme (MLP‐I), a common outer‐layer polyphase precoding matrix is employed. In the second MLP scheme (MLP‐II), data stream corresponding to every transmit antenna is precoded with a different outer‐layer polyphase precoding matrix. Both outer‐layer precoders are custom designed using the trellis exploration algorithm by applying the aperiodic autocorrelation of OFDM data symbols as the metric to minimize. Simulation results indicate that both MLP schemes show superior PAPR performance over conventional MIMO‐OFDM with and without precoding. In addition, MLP better exploits frequency diversity resulting in BER performance gains in multi‐path environments. Copyright © 2010 John Wiley & Sons, Ltd.