基于改进判别字典学习的故障诊断方法

近年来,基于稀疏表示的分类技术在模式识别中取得一定的成功。该框架中,字典的学习和分类器的训练通常是两个独立的模块,降低了方法的识别精度。针对以上问题,提出了一种特征提取和模式识别相融合的改进判别字典学习模型,将重构误差项、稀疏编码判别项及分类误差项进行了整合,并用K奇异值分解算法对目标函数进行优化,实现了字典和分类器的同步学习。该方法先对原始信号进行经验模态分解,并从分解的本征模态函数中提取时、频特征,形成故障样本;然后将训练样本输入改进模型用K奇异值分解优化;最后用习得字典及分类器权重对测试样本进行识别。实验结果表明：该算法不但适用于小样本故障问题,而且鲁棒性和分类性能都明显高于其它算法。         

基于字典学习的主动声呐目标分类方法

主动声呐目标分类在军事和民用方面都有重要的应用和价值。文章基于稀疏表示理论,结合K-奇异值分解和正交匹配追踪算法,提出一种基于学习字典的稀疏表示分类方法（Dictionary Learning Sparse Representation Classification,DLSRC）。首先,利用K-奇异值分解算法训练各个类别目标回波信号,得到带有目标特征信息的类别字典,类别字典对信号具有良好表征能力并且带有目标类别信息;然后,利用正交匹配追踪算法和各个类别字典稀疏分解测试信号,得到各个类别字典下的稀疏系数后重构信号;最后,根据各个重构信号与测试信号的匹配度判定类别,得到分类准确率。结果显示,200个测试数据在信噪比分别为-5、-3、6 dB时,DLSRC法的分类准确率分别达到87%、89%、95.5%。不同信噪比下基于学习字典稀疏表示分类方法的准确率均高于已有的支持向量机（Support Vector Machine,SVM）、K-最近邻（K-Nearest Neighbor,KNN）和柔性最大值分类器（SoftMax）等分类方法,具有较好的分类性能。     

Human Action Recognition Based on Supervised Class-Specific Dictionary Learning with Deep Convolutional Neural Network Features

Human action recognition under complex environment is a challenging work. Recently, sparse representation has achieved excellent results of dealing with human action recognition problem under different conditions. The main idea of sparse representation classification is to construct a general classification scheme where the training samples of each class can be considered as the dictionary to express the query class, and the minimal reconstruction error indicates its corresponding class. However, how to learn a discriminative dictionary is still a difficult work. In this work, we make two contributions. First, we build a new and robust human action recognition framework by combining one modified sparse classification model and deep convolutional neural network (CNN) features. Secondly, we construct a novel classification model which consists of the representation-constrained term and the coefficients incoherence term. Experimental results on benchmark datasets show that our modified model can obtain competitive results in comparison to other state-of-the-art models.     

共享空间基-逐类剩余空间基混合稀疏表示人脸识别算法

针对传统训练样本字典学习未利用类共有信息的不足,引入共享空间和与类别相关的剩余空间,提出了共享空间基-逐类剩余空间基混合稀疏表示人脸识别的算法。该算法首先提取训练样本主成分分析(PCA)特征,获取无标记的共享空间基及其重构样本得到类共有信息;然后结合原始样本得到差分训练集合,并引入类间差异信息构建逐类特异性剩余空间基;最后融合共享空间基和剩余空间基,利用残差判别函数完成模式分类。该方法不仅利用混合空间的正交特性,而且发挥剩余空间的鉴别能力和共享信息稀疏逼近的作用,使结构性字典和模式分类紧密结合。该方法的有效性,分别通过用AR、CMU PIE、Extended Yale B人脸数据库进行的实验得到验证。     

应用一种多核稀疏表示模型实现掌纹分类

与稀疏表示(SR)模型相比,基于单个核函数的SR(KSR)模型可以有效减少数据维数、降低学习模型的计算复杂度并提高特征分类精度;但这种模型对核函数及其参数的选择通常不能包含恰当的、完整的分类信息。为了满足更高的特征分类精度需求,提出了一种基于多个核函数的SR(M-KSR)模型及其快速稀疏优化方法,并将其应用于掌纹图像的分类。测试结果证明了基于M-KSR模型的掌纹分类方法的有效性和实用性。     

利用拼音特征的深度学习文本分类模型

针对人-机器人语音交互中经过语音识别的文本指令,提出了一种利用汉语拼音中声韵母作为特征的深度学习文本分类模型。首先,以无人驾驶车语音导航控制为人机交互的应用背景,分析其文本指令结构并分别构建单一意图与复杂意图语料库;其次,在以字符作为文本分类特征的基础上,结合汉语拼音与英文单词的区别,提出了一种利用拼音声韵母字符作为中文文本分类的特征表示方法;然后,用门控递归单元(GRU)代替传统递归神经网络单元以解决其难以捕获长时间维度特征的不足,为提取信息的高阶特征、缩短特征序列长度并加快模型收敛速度,建立了一种结合卷积神经网络及GRU递归神经网络的深度学习文本分类模型。最后,为验证模型在处理长、短序列任务上的表现,在上述两个语料库上对提出的模型分别进行十折交叉测试,并与其他分类方法进行比较与分析,结果表明该模型显著地提高了分类准确率。     

Weighted residual fusion with multi-modality features for high-resolution aerial scene classification

In scene-level classification of remote sensing, fusion of multi-feature can significantly boost the performance. However, most methods directly fuse the features of different modalities without considering the importance of each feature modality. Based on the above considerations, in this work, multi-modality features weighted residual fusion method is proposed. First, the extracted high-level and low-level features of the scene image are encoded into a unified feature representation. Then the reconstruction residuals of each modality of each scene class are calculated based on two representation-based classification, i.e. sparse representation (SR) and collaborative representation (CR). After fusing the weighted reconstruction residuals of these two modalities with SR and CR, the class label is assigned to the category with the smallest residual. We make extensive evaluations on two challenging remote sensing data sets. The comparison with the state-of-the-art methods demonstrates the effectiveness of our proposed method.     

Weighted Sparse Image Classification Based on Low Rank Representation

The conventional sparse representation-based image classification usually codes the samples independently, which will ignore the correlation information existed in the data. Hence, if we can explore the correlation information hidden in the data, the classification result will be improved significantly. To this end, in this paper, a novel weighted supervised spare coding method is proposed to address the image classification problem. The proposed method firstly explores the structural information sufficiently hidden in the data based on the low rank representation. And then, it introduced the extracted structural information to a novel weighted sparse representation model to code the samples in a supervised way. Experimental results show that the proposed method is superiority to many conventional image classification methods.     

A novel region-based multimodal image fusion technique using improved dictionary learning

Recently, the sparse representation (SR) based algorithms have gained much attention from the researchers in the area of image fusion (IF). The building of a compact discriminative dictionary plays a vital role in the sparse-based IF techniques. In this context, an efficient multimodal IF method based on improved dictionary learning is investigated. The key contributions of this paper are: (a) An improved KSVD algorithm is suggested for the dictionary learning process, (b) to reduce the computational time, only the informative patches are selected using energy feature, and (c) a novel region-based fusion scheme is suggested for the first time for the problem on hand. The suggested technique is tested with a number of multimodal images from Harvard Medical School brain database. The results are compared with state-of-the-art multiscale transform-based methods and modified SR-based methods. Unlike earlier methods, our proposed technique generates an adaptive dictionary through selection of informative patches only. This results in a compact dictionary with improved computational efficiency. The experimental results reveal that our approach outperforms other methods. The potential application of the suggested method could be in pathological images for follow-up study and better treatment planning.     

Review of Text Classification Methods on Deep Learning

Text classification has always been an increasingly crucial topic in natural language processing. Traditional text classification methods based on machine learning have many disadvantages such as dimension explosion, data sparsity, limited generalization ability and so on. Based on deep learning text classification, this paper presents an extensive study on the text classification models including Convolutional Neural Network-Based (CNN-Based), Recurrent Neural Network-Based (RNN-based), Attention Mechanisms-Based and so on. Many studies have proved that text classification methods based on deep learning outperform the traditional methods when processing large-scale and complex datasets. The main reasons are text classification methods based on deep learning can avoid cumbersome feature extraction process and have higher prediction accuracy for a large set of unstructured data. In this paper, we also summarize the shortcomings of traditional text classification methods and introduce the text classification process based on deep learning including text preprocessing, distributed representation of text, text classification model construction based on deep learning and performance evaluation.     

融合2D-PCA及稀疏表示的掌纹识别方法

提出一种基于稀疏表示的掌纹识别方法,该方法借鉴二维主成分分析(PCA)良好的数据压缩属性和较快的特征提取速度,生成掌纹特征图像.二维PCA不仅克服了一维PCA数据维数过大不易计算的缺点,而且保留了原始图像的数据结构,提取的特征能更好的代表原始图像.为了便于稀疏表达,对提取的掌纹特征图像利用一维主成分分析进行二次特征提取,得到训练样本.虽然此处使用了一维PCA,但是由于这是二次特征提取,提取的特征还是保留了原始图像的数据结构,相比单纯的一维PCA,提高了识别率.利用训练样本构造出冗余字典,并采用稀疏表示理论将测试样本表示为字典原子的线性组合,然后根据表示系数的稀疏性与稀疏集中度实现分类识别.由于该方法利用了表达系数的稀疏性,因此减小了算法的时间和空间复杂度.实验表明,针对香港理工大学的MSpalmprints Database,本文方法的识别率较传统方法有明显提高.     

基于核竞争学习算法的图像特征提取

利用核函数学习可有效解决图像特征线性不可分的特性,结合稀疏表示算法的优势,提出了一种新的图像特征提取方法。采用基于竞争学习规则的独立分量分析法对图像进行稀疏表示,该算法可提取数据的高维特征,且不需要优化高阶的非线性函数和进行稀疏密度估计,因而有较快的收敛速度。与仅使用基于竞争学习的独立分量分析法相比,在PolyU数据库上的实验结果表明,采用基于核函数学习和稀疏表示相结合的方法所提取的数据特征有利于提高特征分类精度。     

A novel multimodal medical image fusion using sparse representation and modified spatial frequency

Fusion of multimodal imaging data supports medical experts with ample information for better disease diagnosis and further clinical investigations. Recently, sparse representation (SR)‐based fusion algorithms has been gaining importance for their high performance. Building a compact, discriminative dictionary with reduced computational effort is a major challenge to these algorithms. Addressing this key issue, we propose an adaptive dictionary learning approach for fusion of multimodal medical images. The proposed approach consists of three steps. First, zero informative patches of source images are discarded by variance computation. Second, the structural information of remaining image patches is evaluated using modified spatial frequency (MSF). Finally, a selection rule is employed to separate the useful informative patches of source images for dictionary learning. At the fusion step, batch‐OMP algorithm is utilized to estimate the sparse coefficients. A novel fusion rule which measures the activity level in both spatial domain and transform domain is adopted to reconstruct the fused image with the sparse vectors and trained dictionary. Experimental results of various medical image pairs and clinical data sets reveal that the proposed fusion algorithm gives better visual quality and competes with existing methodologies both visually and quantitatively.     

基于判别字典在线学习的视觉跟踪

提出了一种基于判别字典在线学习的跟踪算法,通过将字典项与标签信息相结合,分类的字典既具有重构性,又具有鉴别性。为了增强模型判别能力,将分类器嵌入到目标表示模型中,依据重构误差和判别分类得分最终确定候选目标。字典学习阶段采用在线字典学习算法同时对字典和分类器进行更新,使模型能够适应目标外观和背景环境的动态变化。实验结果表明,该方法在大量遮挡、快速运动、强光和姿态变化的大部分测试中达到了比较满意的效果。     

基于稀疏非负矩阵分解的低空声目标识别

借鉴人耳听觉原理和特征学习的优势,提出了梅尔(Mel)频率谱提取和稀疏非负矩阵分解相结合的方法用于低空飞行目标声信号识别。首先,以不同目标的Mel频率谱为特征矩阵,利用稀疏非负矩阵分解方法学习得到各自的模板矩阵;然后,利用按列合并后的模板矩阵对训练/测试样本进行特征分解获得编码系数,该系数可作为分类特征;最后,结合不同目标的特点,采用分频段特征提取和顺序二类分类的方法进行多目标分类,并与Mel频率倒谱系数进行性能比较。结果显示,无论在单类目标辨识还是在多类目标分类中,稀疏非负矩阵分解方法均取得了更好的效果。     

An improved SRC method based on virtual samples for face recognition

The sparse representation classifier (SRC) performs classification by evaluating which class leads to the minimum representation error. However, in real world, the number of available training samples is limited due to noise interference, training samples cannot accurately represent the test sample linearly. Therefore, in this paper, we first produce virtual samples by exploiting original training samples at the aim of increasing the number of training samples. Then, we take the intra-class difference as data representation of partial noise, and utilize the intra-class differences and training samples simultaneously to represent the test sample in a linear way according to the theory of SRC algorithm. Using weighted score level fusion, the respective representation scores of the virtual samples and the original training samples are fused together to obtain the final classification results. The experimental results on multiple face databases show that our proposed method has a very satisfactory classification performance.     

Accurate Multi-Scale Feature Fusion CNN for Time Series Classification in Smart Factory

Time series classification (TSC) has attracted various attention in the community of machine learning and data mining and has many successful applications such as fault detection and product identification in the process of building a smart factory. However, it is still challenging for the efficiency and accuracy of classification due to complexity, multi-dimension of time series. This paper presents a new approach for time series classification based on convolutional neural networks (CNN). The proposed method contains three parts: short-time gap feature extraction, multi-scale local feature learning, and global feature learning. In the process of short-time gap feature extraction, large kernel filters are employed to extract the features within the short-time gap from the raw time series. Then, a multi-scale feature extraction technique is applied in the process of multi-scale local feature learning to obtain detailed representations. The global convolution operation with giant stride is to obtain a robust and global feature representation. The comprehension features used for classifying are a fusion of short time gap feature representations, local multi-scale feature representations, and global feature representations. To test the efficiency of the proposed method named multi-scale feature fusion convolutional neural networks (MSFFCNN), we designed, trained MSFFCNN on some public sensors, device, and simulated control time series data sets. The comparative studies indicate our proposed MSFFCNN outperforms other alternatives, and we also provided a detailed analysis of the proposed MSFFCNN.     

Classification of multicolor fluorescence in situ hybridization (M-FISH) images with sparse representation

There has been a considerable interest in sparse representation and compressive sensing in applied mathematics and signal processing in recent years but with limited success to medical image processing. In this paper we developed a sparse representation-based classification (SRC) algorithm based on L1-norm minimization for classifying chromosomes from multicolor fluorescence in situ hybridization (M-FISH) images. The algorithm has been tested on a comprehensive M-FISH database that we established, demonstrating improved performance in classification. When compared with other pixel-wise M-FISH image classifiers such as fuzzy c-means (FCM) clustering algorithms and adaptive fuzzy c-means (AFCM) clustering algorithms that we proposed earlier the current method gave the lowest classification error. In order to evaluate the performance of different SRC for M-FISH imaging analysis, three different sparse representation methods, namely, Homotopy method, Orthogonal Matching Pursuit (OMP), and Least Angle Regression (LARS), were tested and compared. Results from our statistical analysis have shown that Homotopy based method is significantly better than the other two methods. Our work indicates that sparse representations based classifiers with proper models can outperform many existing classifiers for M-FISH classification including those that we proposed before, which can significantly improve the multicolor imaging system for chromosome analysis in cancer and genetic disease diagnosis.     

A supervised dimensionality reduction method-based sparse representation for face recognition

The sparse representation-based classification (SRC) method is a powerful tool to present high-dimensionality data and its superiority in many fields, especially in face recognition application has been proved. With sparsity appropriately harnessed, the SRC can solve face classification problems caused by varying expression, illumination as well as occlusion and disguise. However, face images as high-dimensionality data are usually noisy and the dimensionality is always larger than the number of training sample in real-world applications, which bring a disadvantage for the performance of SRC. Therefore, it is beneficial to perform dimensionality reduction (DR) before utilizing the SRC method. But most prevalent DR methods have no direct connection to SRC. In this paper, we proposed a supervised DR algorithm which suits SRC well and improves the discriminating ability in the low-dimensionality space. The proposed method utilizes the fisher discriminant criterion and low-dimensionality reconstructive restriction to extract the discriminating structure of data. The extensive experiments on public face databases verified the effectiveness of the supervised DR with the model of sparse representation.