BP模式识别与Hopfield联想记忆识别能力的比较

神经网络可以进行模式识别,这一功能具有很高的实用价值.主要研究BP神经网络和离散Hopfield神经网络的识别能力.BP网络是前向神经网络,使用的是监督式学习规则.离散Hopfield是一种反馈网络,结构上输入、输出节点相连使得网络系统的输出状态具有动态变化特性.就数字,字母,手写体数字比较了BP网络和离散Hopfield网络的识别能力,得到了可观的结论,对两个网络的应用做更进一步的阐述.     

基于SVM的车牌数字识别方法

运用支持向量机对车牌字符进行识别,解决了由于图像受客观条件的影响、样本数量不是很大等原因导致的识别率不高的问题.主要针对车牌字符中的数字进行实验,选取了15组数字样本,8组进行训练,7组进行测试,采用交叉验证的思想对SVM进行参数C与g的寻优,并选择合适的核函数,对样本进行训练和预测,对于某些数字的识别率可达到100%,并在相同的训练集和测试集下与BP网络的识别效果进行对比.实验结果表明,SVM在训练样本较少且无字符特征提取的情况下具有很好的识别率,并且有很好的分类推广能力.     

基于剪枝信任度传播的手写体目标形状匹配方法

手写体识别中,目标形状的匹配是较为重要的工作.为了提高手写体目标形状的匹配速度,提出一种新的匹配方法.由于手写体目标形状的几何先验知识已知,并可以采用少量的参数进行表示,新方法采用参数化可变形模板匹配目标形状,确定其后验概率模型,并定义剪枝信任度空间,依据信任度传播算法的特性,首次将剪枝信任度传播算法应用于求解可变形模板与目标形状之间的最佳匹配.实验结果显示,在灰度图像中,对手写体目标形状的轮廓检测与定位速度显著提高.提出将剪枝信任度传播方法应用于手写体目标形状的匹配工作,能够使得目标形状填补空白,应用于相关性较为稀疏的图模型中.     

基于深度学习的验证码识别技术研究

针对基于机器学习的传统验证码识别受字符分割限制与人工操作过多等问题,基于深度学习Tensorflow学习框架将卷积神经网络应用到验证码的特性提取、分析、归类和识别中.将图片验证码作为整体输入,改进传统的LeNet-5网络结构,构建一种端到端的9层卷积神经网络,对验证码图像由低级到高级逐层提取图像特征,实现对图片验证码的识别.模型确定后采用控制变量法,针对每一迭代次数所处理的图片数量进行分析,对其准确率、损失值、训练时间进行综合测评,最终选取最优参数.实验结果显示,每批次处理128张图片,每迭代次数用时6秒,准确率的上限最高达到92%,损失值的下限最低达到0.0184.     

基于ELM与BP神经网络的新疆雹云预测与识别

冰雹是一种强烈的自然灾害,为了提前对冰雹进行预测,提出了一种BP神经网络做预测,极限学习机做识别的预测方法.方法以雷达回波反射率图像的均值、方差、熵为判别指标,首先训练好能够分类云层特征的极限学习机,其次结合雹云形成过程,使用经过较大样本训练的BP神经网络对未来30分钟数据预测,最后使用经过训练的极限学习机进行云层的类型判别.实验结果表明,BP神经网络能够平均提前5-20分钟预测出降雹,具有较好的预测效果.     

基于BP神经网络的仓储火灾火焰识别研究

仓储环境的特殊性限制了传统火灾探测设备的探测效果.为利用现有监控设备,可将图像火灾探测方法引入仓储领域.首先将采集的图像进行预处理,得出感兴趣的目标前景.然后对前景进行特征提取.最后采用基于BP神经网络的识别方法,以提取的图像特征作为输入,对网络进行训练与仿真.实验结果表明:BP神经网络对于火灾火焰图像具有很好的识别能力;作为其输入的图像特征对于火灾火焰图像有着较好的判别效果;为减少硬件投入,采用图像火灾探测方法弥补传统火灾探测设备在仓储应用中的不足是可行的.     

基于PCA与自适应BP算法的侵犯性驾驶行为识别

从机动车安全预警的角度,以有效识别侵犯性驾驶行为为目的,针对传统BP算法学习效率低、收敛速度慢等缺点,提出一种基于PCA与自适应学习速率的BP网络改进识别算法.首先借助人因系统及汽车驾驶平台进行仿真实验,获取驾驶人生理-心理及车辆运行数据集,然后利用主成分分析提取其特征指标,继而应用自适应学习速率BP网络改进算法对驾驶行为进行识别.结果表明:驾驶员呼吸、肌电、速度、油门、车道线偏距以及发动机转速受驾驶行为的影响较大;识别精度为96.17%,对比自适应学习速率BP网络算法、BP网络算法,算法能明显减小训练迭代次数、提高识别精度.     

基于逐步回归分析和BP_Adaboost算法的危险驾驶行为辨识

为准确辨识车辆在行驶过程中可能出现的异常加减速,压线行驶,右侧超车驾驶行为,以便于及时给予驾驶员信息反馈和安全预警,使车辆保持安全的运行状态.首先通过虚拟驾驶仿真实验平台,采集驾驶行为的48种车辆运行数据对实验数据进行预处理,获得实验样本1492组;其次利用逐步回归分析对原始数据进行降维处理,并选取其中的最优回归模型获得特征参数;将提取的特征参数数据输入到BP_Adaboost多分类网络中,训练BP_Adaboost多分类网络,对上述驾驶行为进行识别;最后该模型与BP神经网络进行识别结果对比分析.结果表明模型识别率相较于BP神经网络提高了8.81%,达到92.93%,能进行更加有效的安全预警.     

基于模板匹配的车牌数字与字母识别

车牌的字符识别是车牌识别的重要组成部分.我国的车牌由7个字符组成,除第一位为汉字外,其余均为字母和数字字符,所以提高数字与字母识别率在车牌字符识别中占很重要的地位.通过模板匹配的方法对车牌的数字与字母字符进行识别.选择了两个判别函数,一个是求取模板、待识别字符与运算结果的标准差的最小值,另一个是求取运算结果与模板比值的最大值.并对两个判别函数的识别结果进行了比较,求标准差最小值在总体识别结果上比求比值最大值的识别结果要好,但在个别字符的识别上,求取比值最大值的识别结果要高于求取标准差最小值的识别结果.     

基于MIV-BP型网络实验的房地产项目风险识别研究

为了更准确更客观地识别房地产项目中的风险,为房地产项目投资决策提供科学依据和参考,有效地规避风险,本研究在BP神经网络 (Back-Propagation Neural Network)建模的基础上,采取MIV(Mean Impact Value)算法对BP神经网络模型进行变量筛选的网络优化和改良,从而形成新的优化后的MIV-BP(Mean Impact Value Back-Propagation Neural Network)神经网络,并以此用于评价房地产项目中的风险度以及各因素在风险度中的影响作用大小;同时选取目前相关的房地产项目数据进行仿真实证分析和验证。验证实验结果表明,MIV-BP型神经网络对于房地产项目风险度识别具有良好的适应性和准确性,实验结果客观,达到专家评价的要求,并在风险因素作用度分析上具有良好的应用价值。     

Development of modular neural networks with fuzzy logic response integration for signature recognition

This paper describes a modular neural network (MNN) for the problem of signature recognition. Currently, biometric identification has gained a great deal of research interest within the pattern recognition community. For instance, many attempts have been made in order to automate the process of identifying a person’s handwritten signature, however this problem has proven to be a very difficult task. In this work, we propose an MNN that has three separate modules, each using different image features as input, these are: edges, wavelet coefficients, and the Hough transform matrix. Then, the outputs from each of these modules are combined by using a Sugeno fuzzy integral. The experimental results obtained by using a database of 30 individual’s shows that the modular architecture can achieve a very high 98% recognition accuracy with a test set of 150 images. Therefore, we conclude that the proposed architecture provides a suitable platform to build a signature recognition system.     

A fuzzy rule-based system for handwritten Chinese characters recognition based on radical extraction

In this paper, a fuzzy rule-based system for handwritten Chinese characters recognition (HCCR) based on radical extraction is proposed. Since the writings of handwritten Chinese characters vary a lot, we adopt fuzzy set theory to deal with the recognition of these fuzzy patterns. Candidates of strokes are provided with confidence values to obtain more reliable and accurate results. Furthermore, hierarchical fuzzy rule sets that represent the character structures are used to combine the extracted strokes into compound strokes or radicals. The flexible expansion ability thus provided is very promising. Also, since the number of rules in a fuzzy system is much less than that in a general rule-based system, the computation effort is not difficult. An average of 99.63% recognition rate of 542 test categories that are selected from the 100th sample set of HCCRBASE (character image database provided by CCL, ITRI, Taiwan) is obtained. The experimental results not only verify the feasibility of the proposed system, but also suggest that applying fuzzy set theory to HCCR is an efficient and promising approach.     

数据挖掘在房地产价格预测中的应用研究

通过基于数据挖掘理论的粗糙集和神经网络的研究,用属性约简算法约简并提取了影响房地产价格的主要指标因素,对降维后的数据进行网络学习和训练,最后用训练好的的网络检验测试样本.方法使学习训练的速度和识别率提高了,为房地产价格预测提供了一种更为有效和实用的新途径.     

A novel method for iris feature extraction based on intersecting cortical model network

Iris recognition has received increasing attentions due to its distinct characteristics in recent years. An efficient approach for iris feature extraction plays a very important role in an iris recognition system. In this paper, we developed a novel method for iris feature extraction utilizing the Intersecting Cortical Model (ICM) network which is a simplified model of Pulse-Coupled Neural Network (PCNN) model. In our research, the normalized iris image was imported into an ICM network after enhancement processing. Then the third output pulse image produced by the ICM network was chosen as the iris code. In order to estimate the performance of our iris feature extraction method, an iris recognition platform is produced and the Hamming Distance between two iris codes is computed to measure the dissimilarity of them. The algorithm was tested on CASIA v1.0 iris image database and the results show that the ICM network has promising potential to extract iris textual features.     

BP神经网络在肝硬化治疗预测中的应用

本文利用BP神经网络建立起66例肝硬化治疗结果数据预测模型,并基于matlab得出预测结果。实验证明利用BP神经网络可有效地预测肝病治疗效果。     

MEA-BP模型在遥感影像分类中的应用研究

遥感影像分类作为遥感技术的一个重要应用,对遥感技术的发展具有重要作用.针对遥感影像数据特点,在目前的非线性研究方法中主要用到的是BP神经网络模型.但是BP神经网络模型存在对初始权阈值敏感、易陷入局部极小值和收敛速度慢的问题.因此,为了提高模型遥感影像分类精度,提出采用MEA-BP模型进行遥感影像数据分类.首先采用思维进化算法代替BP神经网络算法进行初始寻优,再用改进BP算法对优化的网络模型权阈值进一步精确优化,随后建立基于思维进化算法的BP神经网络分类模型,并将其应用到遥感影像数据分类研究中.仿真结果表明,新模型有效提高了遥感影像分类准确性,为遥感影像分类提出了一种新的方法,具有广泛研究价值.     

Orbit of an image under iterated system

An orbital picture depicts the path of an object under semi-group of transformations. The concept initially given by Barnsley [3] has utmost importance in image compression, biological modeling and other areas of fractal geometry. In this paper, we introduce superior iterations to study the role of linear and nonlinear transformations on the orbit of an object. Various characteristics of the computed figures have been discussed to indicate the usefulness of study in mathematical analysis. Modified algorithms are given to compute the orbital picture and V-variable orbital picture. An algorithm to calculate the distance between images makes the study motivating. A brief discussion about the proof of the Cauchy sequence of images is also given.     

模糊集在手写数字识别中的应用

为提高识别速度，本文提出先对文字图象进行直（折）线段、圆的ＬＭＳＥ近似并用Ｆｕｚｚｙ定义算法进行描述，以得到对书写工整的部分数字的快速识别，其速度提高近两倍（达到３７５个／秒），约有２０％的字在这一过程中就可以被识别。本文叙述的算法已经用在我们进行的无限制手写数字识别系统的实验中；反复使用表明：本文提出的方法既快又准确。     

Formal description,compression and transformation of digital pictures

In an earlier paper (Part I) we described the construction of Hermite code for multiple grey-level pictures using the concepts of vector spaces over Galois Fields. In this paper a new algebra is worked out for Hermite codes to devise algorithms for various transformations such as translation, reflection, rotation, expansion and replication of the original picture. Also other operations such as concatenation, complementation, superposition, Jordan-sum and selective segmentation are considered. It is shown that the Hermite code of a picture is very powerful and serves as a mathematical signature of the picture. The Hermite code will have extensive applications in picture processing, pattern recognition and artificial intelligence.