Sparse Representation-based Classification Research Articles

Discriminative Dictionary Learning-Based Sparse Classification Framework for Data-Driven Machinery Fault Diagnosis

Data-driven machinery fault diagnosis is important for smart industrial systems to guarantee safety and reliability. However, the conventional data-driven fault diagnosis methods rely on the expert-designed features, which greatly affect the diagnosis performances. Inspired by the sparse representation-based classification (SRC) methods which can learn discriminative sparse features adaptively, we propose a novel discriminative dictionary learning based sparse classification (DDL-SC) framework for data-driven machinery fault diagnosis. The DDL-SC framework can jointly learn a discriminative dictionary for sparse representation and an optimal linear classifier for pattern recognition, which bridges the gaps between two separate processes, dictionary learning and classifier training in traditional SRC methods. In the learning stage, to enhance the discriminability of dictionary learning, we introduce the discriminative sparse code error along with the reconstruction error and classification error into the optimization objective. In the recognition stage, we employ sparse codes of testing signals with respect to the learned discriminative dictionary as inputs of the learned classifier, and promote the recognition performance by connecting a binary hard thresholding operator with the classifier predictions. The effectiveness of DDL-SC is evaluated on the planetary bearing fault dataset and gearbox fault dataset, indicating that DDL-SC yields the recognition accuracies of 99.73% and 99.41%, respectively. Besides, the comparative studies prove the superiority of DDL-SC over several state-of-the-art methods for data-driven machinery fault diagnosis.

Mapping cotton cultivated area combining remote sensing with a fused representation-based classification algorithm

Cotton is one of the major cash crops in the world, which is the main fiber crop that delivers fibers to textile industries across the globe. The efficient and timely information of the cotton distribution is essential for cotton production, agricultural management, as well as the sustainability of agricultural systems. However, there still keep challenges in accurately extracting cotton cultivated area over large areas. To address this issue, we explored the feasibility of combining time series enhanced vegetation index (EVI) calculated from Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data with the fused representation-based classification (FRC) algorithm to identify the cotton pixels and map cotton cultivated area in the major cotton production regions of China. The Fourier transform was used to obtain the harmonic features of the annual time series EVI at a pixel scale over the study area. Then the FRC algorithm was applied for the identification of cotton cultivated area with harmonic features as input. By calculating the fused residual for each pixel, and setting the threshold value for each province based on the statistical data, the cotton cultivated area was extracted in the study area during 2015–2017. Finally, the performance of the FRC algorithm for cotton mapping was evaluated and compared with those of the collaborative representation-based classification (CRC) and sparse representation-based classification (SRC) algorithms. The results showed that there was a good correlation between the MODIS-derived cotton cultivated area and statistical data at the municipal level, with the coefficient of determination (R2) of 0.83 for 3 years. The mapping results of the cotton cultivated area derived by the FRC algorithm had a higher precision compared with those derived by the CRC and SRC algorithms. It demonstrated that the combination of time series MODIS EVI data, statistical data and the FRC algorithm is effective in identifying the cotton cultivated area. Furthermore, the applicability of this approach to other crop mapping and higher spatial resolution images is worth investigating.

Specific Emitter Identification Based on Multi-Level Sparse Representation in Automatic Identification System

Illegally forged signals in automatic identification system (AIS) pose a threat to maritime traffic safety management. In this paper, a multi-level sparse representation based identification (MSRI) algorithm is proposed for specific emitter identification (SEI) in the AIS. The MSRI innovatively combines neural networks with sparse representation based classification (SRC). Channel attention mechanism is introduced to a multi-scale convolutional neural network (CNN) for extracting hidden features in the signal. These extracted features are divided into shallow and deep features according to the depth of the network layer they are extracted from. The original AIS signals and the two-level features are spliced together to form a multi-level dictionary. Subsequently, a sparse representation based identification is performed on the decorrelated multi-level dictionary using the principal components analysis (PCA) method. The proposed MSRI is evaluated on a dataset composed of real-world AIS signals, and compared with the state-of-the-art identification algorithms. The evaluation is based on several factors including computational complexity, number of training samples, and number of emitters. Numerical results indicate that the proposed algorithm can identify emitters with higher accuracy and requires lower training time compared to other methods. Given more than 15 training samples at each emitter, the MSRI can identify nine emitters with an accuracy higher than 90%.

Improving SAR Target Recognition Performance Using Multiple Preprocessing Techniques.

The synthetic aperture radar (SAR) image preprocessing techniques and their impact on target recognition performance are researched. The performance of SAR target recognition is improved by composing a variety of preprocessing techniques. The preprocessing techniques achieve the effects of suppressing background redundancy and enhancing target characteristics by processing the size and gray distribution of the original SAR image, thereby improving the subsequent target recognition performance. In this study, image cropping, target segmentation, and image enhancement algorithms are used to preprocess the original SAR image, and the target recognition performance is effectively improved by combining the above three preprocessing techniques. On the basis of image enhancement, the monogenic signal is used for feature extraction and then the sparse representation-based classification (SRC) is used to complete the decision. The experiments are conveyed on the moving and stationary target acquisition and recognition (MSTAR) dataset, and the results prove that the combination of multiple preprocessing techniques can effectively improve the SAR target recognition performance.

Region Matching of SAR Images Using Blocks for Target Recognition.

A synthetic aperture radar (SAR) target recognition method based on image blocking and matching is proposed. The test SAR image is first separated into four blocks, which are analyzed and matched separately. For each block, the monogenic signal is employed to describe its time-frequency distribution and local details with a feature vector. The sparse representation-based classification (SRC) is used to classify the four monogenic feature vectors and produce the reconstruction error vectors. Afterwards, a random weight matrix with a rich set of weight vectors is used to linearly fuse the feature vectors and all the results are analyzed in a statistical way. Finally, a decision value is designed based on the statistical analysis to determine the target label. The proposed method is tested on the moving and stationary target acquisition and recognition (MSTAR) dataset and the results confirm the validity of the proposed method.

Sparse Representation Classifier Embedding Subspace Mapping and Support Vector for Facial Expression Recognition

With the development of integration and innovation of Internet and industry, facial expression recognition (FER) technology is widely applied in wireless communication and mobile edge computing. The sparse representation‐based classification is a hot topic in computer vision and pattern recognition. It is one type of commonly used image classification algorithms for FER in recent years. To improve the accuracy of FER system, this study proposed a sparse representation classifier embedding subspace mapping and support vector (SRC‐SM‐SV). Based on the traditional sparse representation model, SRC‐SM‐SV maps the training samples into a subspace and extracts rich and discriminative features by using the structural information and label information of the training samples. SRC‐SM‐SV integrates the support vector machine to enhance the classification performance of sparse representation coding. The solution of SRC‐SM‐SV uses an alternate iteration method, which makes the optimization process of the algorithm simple and efficient. Experiments on JAFFE and CK+ datasets prove the effectiveness of SRC‐SM‐SV in FER.

Remote Sensing Scene Classification Using Sparse Representation-Based Framework With Deep Feature Fusion

Scene classification of high-resolution remote sensing (RS) images has attracted increasing attentions due to its vital role in a wide range of applications. Convolutional neural networks (CNNs) have recently been applied on many computer vision tasks and have significantly boosted the performance including imagery scene classification, object detection, and so on. However, the classification performance heavily relies on the features that can accurately represent the scene of images, thus, how to fully explore the feature learning ability of CNNs is of crucial importance for scene classification. Another problem in CNNs is that it requires a large number of labeled samples, which is impractical in RS image processing. To address these problems, a novel sparse representation-based framework for small-sample-size RS scene classification with deep feature fusion is proposed. Specially, multilevel features are first extracted from different layers of CNNs to fully exploit the feature learning ability of CNNs. Note that the existing well-trained CNNs, e.g., AlexNet, VGGNet, and ResNet50, are used for feature extraction, in which no labeled samples is required. Then, sparse representation-based classification is designed to fuse the multilevel features, which is especially effective when only a small number of training samples are available. Experimental results over two benchmark datasets, e.g., UC-Merced and WHU-RS19, demonstrated that the proposed method can effectively fuse different levels of features learned in CNNs, and clearly outperform several state-of-the-art methods especially with limited training samples.

Combination of multiple decision principles based on sparse representation-based classification for target recognition of SAR image

A synthetic aperture radar (SAR) target recognition method using multiple decision principles in sparse representation-based classification (SRC) was proposed. The traditional SRC generally reconstructed the test sample on the global dictionary and calculated the reconstruction errors of individual training classes. And the decision was reached based on the minimum reconstruction error. However, because of the complexity of SAR target recognition, a single decision principle probably had low adaptivity to the extended operating conditions (EOC). Therefore, this paper employed the global minimum reconstruction, maximum coefficient energy, and local minimum reconstruction error principles to make decisions based on the solved coefficient vector from sparse representation. The global minimum reconstruction error principle directly adopted the traditional one. The maximum coefficient energy principle calculated the coefficient energies of different classes and made decision based on the maximum one. The local minimum reconstruction error principle represented and analyzed the test sample on the local dictionary so the azimuthal sensitivity of SAR imaging could be exploited. For the decision values from the three principles, they were transformed to the same type of probability vectors. Finally, the linear fusion was performed to combine their decisions. Experiments were conducted on the MSTAR dataset under situations including the standard operating condition (SOC), depression angle variance, noise corruption, and target occlusion. The results validate that the combination of multiple decision principles could effectively improve SAR target recognition performance.

Target Recognition of SAR Images Based on SVM and KSRC.

A synthetic aperture radar (SAR) target recognition method combining linear and nonlinear feature extraction and classifiers is proposed. The principal component analysis (PCA) and kernel PCA (KPCA) are used to extract feature vectors of the original SAR image, respectively, which are classical and reliable feature extraction algorithms. In addition, KPCA can effectively make up for the weak linear description ability of PCA. Afterwards, support vector machine (SVM) and kernel sparse representation-based classification (KSRC) are used to classify the KPCA and PCA feature vectors, respectively. Similar to the idea of feature extraction, KSRC mainly introduces kernel functions to improve the processing and classification capabilities of nonlinear data. Through the combination of linear and nonlinear features and classifiers, the internal data structure of SAR images and the correspondence between test and training samples can be better investigated. In the experiment, the performance of the proposed method is tested based on the MSTAR dataset. The results show the effectiveness and robustness of the proposed method.

Recognition of Cognitive Impairment in Adult Moyamoya Disease: A Classifier Based on High-Order Resting-State Functional Connectivity Network.

Objective: Vascular cognitive impairment (VCI) is a common complication in adult patients with moyamoya disease (MMD), and is reversible by surgical revascularization in its early stage of mild VCI. However, accurate diagnosis of mild VCI is difficult based on neuropsychological examination alone. This study proposed a method of dynamic resting-state functional connectivity (FC) network to recognize global cognitive impairment in MMD.Methods: For MMD, 36 patients with VCI and 43 patients with intact cognition (Non-VCI) were included, as well as 26 normal controls (NCs). Using resting-state fMRI, dynamic low-order FC networks were first constructed with multiple brain regions which were generated through a sliding window approach and correlated in temporal dimension. In order to obtain more information of network interactions along the time, high-order FC networks were established by calculating correlations among each pair of brain regions. Afterwards, a sparse representation-based classifier was constructed to recognize MMD (experiment 1) and its cognitive impairment (experiment 2) with features extracted from both low- and high-order FC networks. Finally, the ten-fold cross-validation strategy was proposed to train and validate the performance of the classifier.Results: The three groups did not differ significantly in demographic features (p > 0.05), while the VCI group exhibited the lowest MMSE scores (p = 0.001). The Non-VCI and NCs groups did not differ significantly in MMSE scores (p = 0.054). As for the classification between MMD and NCs, the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity of the classifier reached 90.70, 88.57, 93.67, and 73.08%, respectively. While for the classification between VCI and Non-VCI, the AUC, accuracy, sensitivity, and specificity of the classifier reached 91.02, 84.81, 80.56, and 88.37%, respectively.Conclusion: This study not only develops a promising classifier to recognize VCI in adult MMD in its early stage, but also implies the significance of time-varying properties in dynamic FC networks.

Face Recognition Based on Deep Features

A face recognition method based on deep feature decision fusion is proposed. First, a convolutional neural network (CNN) is designed for deep feature learning on face images. Then, all the feature are mapped to construct a single depth feature vector. In the classification stage, the sparse representation-based classification is used to characterize the constructed depth feature vector, and the face category of the object to be recognized is determined according to the overall reconstruction error. Experiments are carried out on ORL and Yale-B datasets and compared with several existing face recognition methods. The results verify the effectiveness and robustness of the proposed method.

Generalized and Discriminative Collaborative Representation for Multiclass Classification.

This article presents a generalized collaborative representation-based classification (GCRC) framework, which includes many existing representation-based classification (RC) methods, such as collaborative RC (CRC) and sparse RC (SRC) as special cases. This article also advances the GCRC theory by exploring theoretical conditions on the general regularization matrix. A key drawback of CRC and SRC is that they fail to use the label information of training data and are essentially unsupervised in computing the representation vector. This largely compromises the discriminative ability of the learned representation vector and impedes the classification performance. Guided by the GCRC theory, we propose a novel RC method referred to as discriminative RC (DRC). The proposed DRC method has the following three desirable properties: 1) discriminability: DRC can leverage the label information of training data and is supervised in both representation and classification, thus improving the discriminative ability of the representation vector; 2) efficiency: it has a closed-form solution and is efficient in computing the representation vector and performing classification; and 3) theory: it also has theoretical guarantees for classification. Experimental results on benchmark databases demonstrate both the efficacy and efficiency of DRC for multiclass classification.

Video action recognition with visual privacy protection based on compressed sensing

Action recognition in video involves interpreting complex semantic information, which remains a challenge in computer vision. Research on the use of deep neural networks for video action recognition has made significant progress in recent years, but most available recognition frameworks cannot satisfactorily perform the video action recognition task in privacy scene. In light of this issue, this paper proposes a method for video action recognition that employs a compressed sensing-based visual privacy protection framework. The proposed method protects visual privacy by balancing operational efficiency with the accuracy of recognition. It consists of three steps. First, compressed sensing is used to realize visual privacy protection and significantly enhance efficiency at the same time. Second, the convolutional 3D network model is used due to its speed to represent features of video actions in the compressed sensing data, and PCA is used to reduce the dimensionality of the extracted feature vectors to reduce temporal complexity. Finally, a sparse representation-based classification algorithm is integrated into the AdaBoost architecture to improve the recognition performance of the proposed method. Experiments were performed on three commonly used video action datasets to verify the proposed method, and the results show that it is robust in video action recognition tasks, and adequately protects visual privacy.

Two-stage sparse representation of NSCT features with application to SAR target classification

To improve synthetic aperture radar (SAR) target classification performance, both the feature extraction and classification algorithms are considered. The nonsubsampled contourlet transform (NSCT) is employed to decompose SAR images to generate multi-layer components, which provide more discriminative descriptions for the targets. During the classification, the two-stage sparse representations are developed. Each NSCT component is classified using the sparse representation-based classification (SRC). Afterwards, the joint sparse representation (JSR) is adopted to represent the selected NSCT components from the first stage, which are assumed to be highly discriminative. Based on the two decisions, the target label of the test sample is determined. In the experiment, the moving and stationary target acquisition and recognition (MSTAR) dataset is used to set up scenarios for performance evaluation including the standard operating condition (SOC) and several extended operating conditions (EOCs). The results show the superior performance of the proposed methodover some current methods.

Ensemble sparse representation-based cyber threat hunting for security of smart cities

The ever-growing expansion of smart cities and the Internet of Things (IoT) offer a promising solution to many contemporary urban challenges. However, this digital transformation also results in cyber-security loopholes which can be exploited by malicious hackers to wreak substantial digital and physical damage. Malware is the primary tool of cyber-criminals for attacking digital systems. In this paper, a multi-view ensemble threat hunting model based on Sparse Representation based Classifier (SRC) is proposed to use in IoT systems that are finding domain space in the advent of Smart Cities. An ensemble of SRCs is considered where every individual SRC classifies malware by Opcode, Bytecode and system call views of several standard IoT and Ransomware datasets. The final decision is made through weighted majority voting. SRC is employed to alleviate the complexity of the base classifiers. Experimental results verify the efficiency and robustness of the proposed model in different balanced and imbalanced environments. The proposed model outperforms all base classifiers and several well-known works in current literature.

Image Target Recognition via Mixed Feature-Based Joint Sparse Representation.

An image target recognition approach based on mixed features and adaptive weighted joint sparse representation is proposed in this paper. This method is robust to the illumination variation, deformation, and rotation of the target image. It is a data-lightweight classification framework, which can recognize targets well with few training samples. First, Gabor wavelet transform and convolutional neural network (CNN) are used to extract the Gabor wavelet features and deep features of training samples and test samples, respectively. Then, the contribution weights of the Gabor wavelet feature vector and the deep feature vector are calculated. After adaptive weighted reconstruction, we can form the mixed features and obtain the training sample feature set and test sample feature set. Aiming at the high-dimensional problem of mixed features, we use principal component analysis (PCA) to reduce the dimensions. Lastly, the public features and private features of images are extracted from the training sample feature set so as to construct the joint feature dictionary. Based on joint feature dictionary, the sparse representation based classifier (SRC) is used to recognize the targets. The experiments on different datasets show that this approach is superior to some other advanced methods.

Block Sparse Bayesian Learning over Local Dictionary for Robust SAR Target Recognition

This paper applied block sparse Bayesian learning (BSBL) to synthetic aperture radar (SAR) target recognition. The traditional sparse representation-based classification (SRC) operates on the global dictionary collaborated by different classes. Afterwards, the similarities between the test sample and various classes are evaluated by the reconstruction errors. This paper reconstructs the test sample based on local dictionaries formed by individual classes. Considering the azimuthal sensitivity of SAR images, the linear coefficients on the local dictionary are sparse ones with block structure. Therefore, to solve the sparse coefficients, the BSBL is employed. The proposed method can better exploit the representation capability of each class, thus benefiting the recognition performance. Based on the experimental results on the moving and stationary target acquisition and recognition (MSTAR) dataset, the effectiveness and robustness of the proposed method is confirmed.

Target recognition of SAR images using fused deep feature by multiset canonical correlations analysis

A target recognition method of synthetic aperture radar (SAR) image is developed based on deep features learned from convolutional neural networks (CNN). The learned feature maps by convolution operations are combined as one unified feature vector through multiset canonical correlations analysis (MCCA). Afterwards, the sparse representation-based classification (SRC) is adopted as the classifier for target recognition. The proposed method could effectively combine the discriminability of different feature maps. In addition, the advantages of CNN and SRC can be integrated simultaneously. In the experiments, several conditions are setup based on the moving and stationary target acquisition and recognition (MSTAR) dataset for the validation. According to the results, the proposed method keeps high effectiveness and robustness under standard operating condition (SOC) and extended operating conditions (EOC).

A Hierarchical Discriminative Sparse Representation Classifier for EEG Signal Detection.

Classification of electroencephalogram (EEG) signal data plays a vital role in epilepsy detection. Recently sparse representation-based classification (SRC) methods have achieved the good performance in EEG signal automatic detection, by which the EEG signals are sparsely represented using a few active coefficients in the dictionary and classified according to the reconstruction criteria. However, most of SRC learn a linear dictionary for encoding, and cannot extract enough information and nonlinear relationship of data for classification. To solve this problem, a hierarchical discriminative sparse representation classification model (called HD-SRC) for EEG signal detection is proposed. Based on the framework of neural network, HD-SRC learns the hierarchical nonlinear transformation and maps the signal data into the nonlinear transformed space. Through incorporating this idea into label consistent K singular value decomposition (LC-KSVD) at the top layer of neural network, HD-SRC seeks discriminative representation together with dictionary, while minimizing errors of classification, reconstruction and discriminative sparse-code for pattern classification. By learning the hierarchical feature mapping and discriminative dictionary simultaneously, more discriminative information of data can be exploited. In the experiment the proposed model is evaluated on the Bonn EEG database, and the results show it obtains satisfactory classification performance in multiple EEG signal detection tasks.

On-line detection of porosity change of high temperature blade coating for gas turbine

The health status of high-temperature blades is one of the key factors that affect the operation of the gas turbine. Since turbine blades usually work in high temperature and high pressure environments, the thermal insulation performance of Thermal barrier coating (TBC) is the main factor to determine the service life of high-temperature blades. The change of porosity of TBC has a great influence on its thermal insulation performance. In order to establish an on-line monitoring system for thermal insulation performance of high temperature blade TBC, it is necessary to obtain the microstructure evolution law of TBC in real time. In this work, a coupling algorithm based on Gray gradient space histogram entropy (GGSHE) and Sparse representation-based classifier (SRC) was developed. The investigation of this Non-destructive testing (NDT) method is important to evaluate the variation of the porosity of TBC with the service time by analyzing the image of the surface temperature distribution of the coating. The effectiveness of GGSHE is trained and verified by using the numerical data, and compared with other algorithms. The results show that the GGSHE has the best detection performance: the Average error rate (AER) of GGSHE is 3.23% under the constant working condition, and 4.18% under the changing working condition, which are both the best results. Moreover, when the microstructure of coating changes or the temperature legend of infrared thermal imaging becomes larger, GGSHE can still extract the feature of pores effectively and can achieve high detection accuracy. The detection time of GGSHE is 4.04 s, meeting the time requirements of on-line detection of high temperature blades.