Widrow-Hoff神经网络学习规则的应用研究

基于线性神经网络原理,提出线性神经网络的模型,并利用Matlab实现Widrow-Hoff神经网络算法。分析Matlab人工神经网络工具箱中有关线性神经网络的工具函数,最后给出线性神经网络在系统辨识中的实际应用。通过对线性神经网络的训练,进一步验证Widrow-Hoff神经网络算法的有效性,以及用其进行系统辨识的高精度拟合性。     

Widrow-Hoff神经网络学习规则的应用研究

基于线性神经网络原理,提出线性神经网络的模型,并利用Matlab实现Widrow-Hoff神经网络算法.分析Matlab人工神经网络工具箱中有关线性神经网络的工具函数,最后给出线性神经网络在系统辨识中的实际应用.通过对线性神经网络的训练,进一步验证Widrow-Hoff神经网络算法的有效性,以及用其进行系统辨识的高精度拟合性.     

毫米波混合波束形成系统的多波束训练方法

针对毫米波通信系统方向性强的特点,提出一种快速波束训练方法。该方法首先建立了二维面阵天线系统模型,推导了多波束预编码矩阵,并在原分层训练协议的基础上,利用混合波束形成系统使发射端同时产生多个并行训练波束并在接收端将训练结果进行反馈。仿真表明,相较于以往单一波束训练算法,该并行多波束训练方法能够有效地减少训练开销,提高训练效率。     

一种快速多层神经网络训练算法及其在系统辩识中的应用

本文提出了一种新的多层前馈神经网络快速训练方法.该算法是基于指数加权局部最小二乘(EWLLS)目标函数及殴几里得方向集(EDS)方法的,在训练过程中,通过估计局部期望输出,多层神经网络可以被分解成若干个自适应线性神经元(Adaline),而Adaline是通过EDS方法进行训练的.该算法的性能是通过将其应用于系统辩识中加以说明的.     

一种基于逐维扩展的前向网络训练方法

前向网络的快速训练问题是前向网络研究的一个非常重要的课题。本文针对一类n-维超立方体的分类问题（当为二分类问题时，这实际上是一个n-维Boole函数的神经网络实现问题），提出了一种基于逐维扩展的前向网络快速训练方法，将一个n个输入的大网络的各权训练问题转化为小网络逐维递归的扩展部分的参数训练问题，提高了网络训练的速度，实验结果表明了这种训练方法的有效性和可行性。     

GMD-SDDBHMM语音识别模型和分类训练方法

本文将混合高斯分布应用于一种非齐次隐含马尔可夫模型——简化的基于段长分布的隐含马尔可夫模型。新模型使语音识别率得到了改善。由于通常的模型训练方法训练时间太长，本文提出了一种分类训练方法，在不降低最终模型性能的前提下，使训练可以分布式完成。     

语音识别模型和分类训练方法

本文将混合高斯分布应用于一种非齐次隐含马尔可夫模型——简化的基于段长分布的隐含马尔可夫模型。新模型使语音识别率得到了改善。由于通常的模型训练方法训练时间太长，本文提出了一种分类训练方法，在不降低最终模型性能的前提下，使训练可以分布式完成。     

电动式扬声器单元支撑系统的蠕变效应

电动式扬声器单元支撑系统的蠕变效应表现在扬声器单元的位移在共振频率以下会有所上升.因此在扬声器单元线性集中参数模型中仅仅用线性弹簧并联简单的粘滞阻尼来代表支撑系统是不够的.在此引入对数蠕变模型,通过测量实际的扬声器单元与传统的线性模型进行对比,结果表明对数蠕变模型与实际测量结果更吻合.     

基于遗传算法的RBF神经网络在涡流传感器非线性补偿中应用

为了解决涡流传感器的非线性问题,应用遗传算法(GA)训练径向基函数(RBF)神经网络(NN)实现其非线性补偿.介绍非线性补偿的原理和网络训练方法.从实测数据出发,建立了涡流传感器的非线性补偿模型.该方法能同时优化网络结构和参数,具有全局寻优能力,补偿精度高、鲁棒性好、网络训练速度快、能实现在线软补偿.实验结果表明,所采用的涡流传感器非线性补偿方法是有效的和可行的.补偿后,最大非线性误差在0.5%范围内,具有良好的线性.     

基于灾难性遗忘及组合叠加擦除的跨模态行人重识别预训练方法

面向构建24小时全时段视频监控系统的需要，基于可见光与近红外的跨模态行人重识别受到工业界与学术界的广泛关注.然而，目前大部分跨模态行人重识别任务都试图利用在ImageNet上预训练的模型来提前学习模态内共性特征，但ImageNet与跨模态行人数据模态差异较大，且预训练过程中将颜色信息作为判别特征之一，导致预训练中学习到的共性特征并不适用于无色彩红外图像的信息表示.本文提出了一种基于灾难性遗忘及组合叠加擦除的自监督跨模态行人重识别预训练方法，首先利用提出的灾难性遗忘评分来对预训练数据进行筛选，旨在减小预训练数据与后续任务数据存在的域间差距，进一步减少模型训练时间.其次，针对传统跨模态识别中的关键区分性特征提取，本文设计了一种强通道数据增强策略，通过对R、G、B三通道的通道级擦除与组合，生成了颜色迥异的多类型样本，有利于促使模型关注于纹理信息而非颜色信息.最后基于本文提出的跨模态数据筛选指标以及通道增强策略，构建了跨模态任务的自监督学习框架.实验结果表明，本文提出的预训练方法所训练的ResNet50网络在迁移到众多跨模态行人重识别方法时优于目前主流自监督预训练方法，其中在经典方法 AGW的...     

Hybrid IMM/SVM approach for wavelet-domain probabilistic model based texture classification

The Fisher kernel method was recently proposed to incorporate probabilistic (generative) models and discriminative methods for pattern recognition. This method uses parameter derivatives of log-likelihood calculated from probabilistic model(s), Fisher scores, to generate statistical feature vectors. It is followed by discriminative classifiers such as the support vector machine (SVM) for classification. In this work, the authors study the potential of the Fisher kernel method on texture classification. A hybrid system of independent mixture model (IMM) and SVM is introduced to extract and classify statistical texture features in the wavelet-domain. Compared to existing methods that apply Bayesian classification based on wavelet domain energy signatures and stand alone IMM, the new hybrid IMM/SVM method is able to achieve superior performance. Experimental results are presented to demonstrate the effectiveness of this proposed method.     

Collaborative object tracking model with local sparse representation

There existed many visual tracking methods that are based on sparse representation model, most of them were either generative or discriminative, which made object tracking more difficult when objects have undergone large pose change, illumination variation or partial occlusion. To address this issue, in this paper we propose a collaborative object tracking model with local sparse representation. The key idea of our method is to develop a local sparse representation-based discriminative model (SRDM) and a local sparse representation-based generative model (SRGM). In the SRDM module, the appearance of a target is modeled by local sparse codes that can be formed as training data for a linear classifier to discriminate the target from the background. In the SRGM module, the appearance of the target is represented by sparse coding histogram and a sparse coding-based similarity measure is applied to compute the distance between histograms of a target candidate and the target template. Finally, a collaborative similarity measure is proposed for measuring the difference of the two models, and then the corresponding likelihood of the target candidates is input into a particle filter framework to estimate the target state sequentially over time in visual tracking. Experiments on some publicly available benchmarks of video sequences showed that our proposed tracker is robust and effective.     

Robust discriminative extreme learning machine for relevance feedback in image retrieval

Relevance feedback (RF) has long been an important approach for multi-media retrieval because of the semantic gap in image content, where SVM based methods are widely applied to RF of content-based image retrieval. However, RF based on SVM still has some limitations: (1) the high dimension of image features always make the RF time-consuming; (2) the model of SVM is not discriminative, because labels of image features are not sufficiently exploited. To solve above problems, we proposed robust discriminative extreme learning machine (RDELM) in this paper. RDELM involved both robust within-class and between-class scatter matrices to enhance the discrimination capacity of ELM for RF. Furthermore, an angle criterion dimensionality reduction method is utilized to extract the discriminative information for RDELM. Experimental results on four benchmark datasets (Corel-1K, Corel-5K, Corel-10K and MSRC) illustrate that our proposed RF method in this paper achieves better performance than several state-of-the-art methods.     

Histopathological image classification through discriminative feature learning and mutual information-based multi-channel joint sparse representation

Histopathological image classification is a very challenging task because of the biological heterogeneities and rich geometrical structures. In this paper, we propose a novel histopathological image classification framework, which includes the discriminative feature learning and the mutual information-based multi-channel joint sparse representation. We first propose a stack-based discriminative prediction sparse decomposition (SDPSD) model by incorporating the class labels information to predict deep discriminant features automatically. Subsequently, a mutual information-based multi-channel joint sparse model (MIMCJSM) is presented to jointly encode the common component and particular components of the discriminative features. Especially, the main advantage of the MIMCJSM is the construction of a joint dictionary using a mutual information criterion, which contains a common sub-dictionary and three particular sub-dictionaries. Based on the joint dictionary, the MIMCJSM captures the relationship of multi-channel features, which can improve discriminative ability of joint sparse representation coefficients. Finally, the joint sparse representation coefficients of different levels can be aggregated using the spatial pyramid matching (SPM) model, and the linear support vector machine (SVM) is used as the classifier. Experimental results on ADL and BreaKHis datasets demonstrate that our proposed framework consistently performs better than popular existing classification frameworks. Additionally, it can show promising strong-robustness performance for histopathological image classification.     

基于支持向量机的贯流风叶叶片粘连缺陷诊断

为了提高贯流风叶叶片粘连缺陷诊断的准确率和鲁棒性,提出了一种基于支持向量机的贯流风叶叶片粘连缺陷诊断方法。该方法以线性核函数为内积核函数,在追求分类间隔最大化的前提下,建立了叶片粘连缺陷诊断数学模型。仿真和实际测试结果表明,即使在使用较少的训练样本的情况下,该模型仍能达到较高的叶片粘连缺陷诊断率,效果优于传统的诊断方法,为贯流风叶叶片粘连缺陷诊断提供了新的途径。     

一种改进的GOP算法在区分性训练的应用

自动发音错误检错中基于最大化F1值的区分性训练方法是最近提出来的一种声学模型训练方法,该方法能够有效增大发音检错系统中的训练和测试数据检错的Fl值。对发音质量评估方法上进行研究,提出一种改进的GOP算法来替代传统的GOP算法,改进GOP算法把传统地GOP算法的先求后验概率再求时间归一化改变成先求时间归一化再求后验概率。根据改进GOP算法给出了使用改进GOP算法最大F1准则的参数更新公式,发音检错实验结果表明基于改进的GOP算法的最大F1值准则训练较使用传统的GOP算法具有过训练抑制性好,在训练机上较低的目标函数值上能达到较高的测试集上的F1值等较好的性能。     

基于支持向量机的说话人辨认研究

支持向量机是统计学理论的一个重要的学习方法，也是解决模式识别问题的一个有力的工具，本文提出了用支持向量机来解决说话人辨认问题。结合语音信号的特点，解决了大数据量情况下支持向量机的训练问题。支持向量机对两类的分类问题有着突出的优势，本文用两种判决规则将两类问题应用到多类的识别问题。用支持向量机实现了一个与文本无关的说话人辨认系统，实验表明，本方法有良好的效果。     

Robust discriminative broad learning system for hyperspectral image classification

With the advantages of simple structure and fast training speed, broad learning system (BLS) has attracted attention in hyperspectral images (HSIs). However, BLS cannot make good use of the discriminative information contained in HSI, which limits the classification performance of BLS. In this paper, we propose a robust discriminative broad learning system (RDBLS). For the HSI classification, RDBLS introduces the total scatter matrix to construct a new loss function to participate in the training of BLS, and at the same time minimizes the feature distance within a class and maximizes the feature distance between classes, so as to improve the discriminative ability of BLS features. RDBLS inherits the advantages of the BLS, and to a certain extent, it solves the problem of insufficient learning in the limited HSI samples. The classification results of RDBLS are verified on three HSI datasets and are superior to other comparison methods.     

基于主动标记支持向量机和太赫兹光谱的转基因物质检测方法研究

为克服传统支持向量机需要事先对训练样本进行 人为标记的缺点,提出了一种主动训练 支持向量机模型。利用仿射传播聚类算法对未标记样本进行聚类分析,在迭代过程中 不断更新现 有支持向量机的训练数据,从而不仅可以减少人为标记样本所带来的误差,还能够最大限度 地提高模型的识 别准确率。本文以转基因棉花的太赫兹光谱数据为研究对象对该模型进行了验证,实验结果 表明,本文 提出的方法对总待测样品的种类的识别率为95.56%,较其他三种方法 有较少的误判和更高的识别率。 基于仿射传播聚类的支持向量机较传统支持向量机有更高的识别率和更低的误判率,为转基 因物质的检测提供了一种快速,无损的新方法。     

Multi-level feature fusion and multi-loss learning for person Re-Identification

With the rise of deep learning technology, person re-identification (Re-id) technology has been developed rapidly. During the training process, many recent methods are susceptible to target misalignment and without sufficient discriminative features. Aiming at these two problems, a simple and potent model is proposed by us. A new self-attention module and a multi-loss function with relative weight are designed to integrate the multi-level features of pedestrians in our network. Specifically, the goal of the self-attention is to instruct the baseline network to learn robust features from the resized images. The key of the self-attention model is the weighting of the importance of different person regions. Therefore, the non-local features of the different levels in the baseline network will be paid more attention, which is conducive to learn discriminative features from the baseline network. Finally, a multi-loss function with relative weight is introduced to enhance the feature learning ability and integrate more features reasonably. Many experiments have been done on the three datasets (Market1501, DukeMTMC-reID and CUHK03-NP) and the results explain that the new model gets a higher accuracy than many other recent approaches.