Search

Available data in machine learning applications is becoming increasingly complex, due to higher dimensionality and difficult classes. There exists a wide variety of approaches to measuring complexity of labeled data, according to class overlap, separability or boundary shapes, as well as group morphology. Many techniques can transform the data in order to find better features, but few focus on specifically reducing data complexity. Most data transformation methods mainly treat the dimensionality aspect, leaving aside the available information within class labels which can be useful when classes are somehow complex. This paper proposes an autoencoder-based approach to complexity reduction, using class labels in order to inform the loss function about the adequacy of the generated variables. This leads to three different new feature learners, Scorer, Skaler and Slicer. They are based on Fisher’s discriminant ratio, the Kullback-Leibler divergence and least-squares support vector machines, respectively. They can be applied as a preprocessing stage for a binary classification problem. A thorough experimentation across a collection of 27 datasets and a range of complexity and classification metrics shows that class-informed autoencoders perform better than 4 other popular unsupervised feature extraction techniques, especially when the final objective is using the data for a classification task. [ABSTRACT FROM AUTHOR]

Available data in machine learning applications is becoming increasingly complex, due to higher dimensionality and difficult classes. There exists a wide variety of approaches to measuring complexity of labeled data, according to class overlap, separability or boundary shapes, as well as group morphology. Many techniques can transform the data in order to find better features, but few focus on specifically reducing data complexity. Most data transformation methods mainly treat the dimensionality aspect, leaving aside the available information within class labels which can be useful when classes are somehow complex. This paper proposes an autoencoder-based approach to complexity reduction, using class labels in order to inform the loss function about the adequacy of the generated variables. This leads to three different new feature learners, Scorer, Skaler and Slicer. They are based on Fisher’s discriminant ratio, the Kullback-Leibler divergence and least-squares support vector machines, respectively. They can be applied as a preprocessing stage for a binary classification problem. A thorough experimentation across a collection of 27 datasets and a range of complexity and classification metrics shows that class-informed autoencoders perform better than 4 other popular unsupervised feature extraction techniques, especially when the final objective is using the data for a classification task. [ABSTRACT FROM AUTHOR]

In this article, a modified convolutional neural network (CNN) algorithm, namely 1D-GAPCNN-SVM, is proposed to address the early anomaly diagnosis problem. Considering the fact that traditional 2D-CNN based approaches contain too many model parameters and are not suitable for fast diagnosis applications using multisensor 1-D time-series measurements, 1D-CNN is introduced to deal with this problem. To reduce the number of parameters, a 1-D global average pooling layer is designed to substitute the fully connected layer with two or three layers. In order to further improve the diagnosis accuracy, a nonlinear multiclass support vector machine (SVM) is adopted to replace the traditional Softmax classifier as the final discriminator. Raw multisensor 1-D time-series data are directly fed into the diagnosis model, then the diagnosis result can be automatically generated. Two experiments, which are rolling bearing and GPS anomaly detection, have been conducted to demonstrate the effectiveness and the superior performance of the proposed method compared to the conventional SVM, K-nearest neighbor, deep neural network (DNN), and traditional 2D-CNN. [ABSTRACT FROM AUTHOR]

Android is a growing target for malicious software (malware) because of its popularity and functionality. Malware poses a serious threat to users’ privacy, money, equipment and file integrity. A series of data-driven malware detection methods were proposed. However, there exist two key challenges for these methods: (1) how to learn effective feature representation from raw data; (2) how to reduce the dependence on the prior knowledge or human labors in feature learning. Inspired by the success of deep learning methods in the feature representation learning community, we propose a malware detection framework which starts with learning rich-features by a novel unsupervised feature learning algorithm Merged Sparse Auto-Encoder (MSAE). In order to extract more compact and discriminative feature from the rich-features to further boost the malware detection capability, a hybrid deep network learning algorithm Stacked Hybrid Learning MSAE and SDAE (SHLMD) is established by further incorporating a classical deep learning method Stacked Denoising Auto-encoders (SDAE). After that, we feed the feature learned by MSAE and SHLMD respectively to classification algorithms, e.g., Support Vector Machine (SVM) or K-NearestNeighbor (KNN), to train a malware detection model. Evaluation results on two real-world datasets demonstrate that SHLMD achieves 94.46 and 90.57 percent accuracy respectively, which outperforms the classical unsupervised feature representation learning Sparse Auto-encoder (SAE). MSAE performs similarly to SAE. SHLMD can further improve the performance of MSAE and the supervised fine-tuned method SDAE. Besides, we compare the performance of our methods with that of state-of-the-art detection approaches, including classical deep-learning-based methods. Extensive experiments show that our proposed methods are effective enough to detect Android malware. [ABSTRACT FROM AUTHOR]

The natural gas tightly correlates with our everyday life. However, driven by gray incomes, some users are prone to stealing gas by refitting the equipment without permission. Especially for the boiler room users in winter, this phenomenon appears more rampant. Traditional gas-theft detection methods highly rely on the on-site inspection, where exists ineffective and randomness. With the rapidly deployed IoT sensors, we can collect real-time gas consumption data to analyze users’ behavior patterns, where the gas-theft suspects could be discovered early and accurately. In this paper, we propose a data-driven approach, named SVOC, to detect gas-theft suspects among boiler room users. Our approach consists of a scenario-based data quality detection algorithm, a deformation-based normality detection algorithm, and an One-Class Support Vector Machine (OCSVM) based anomaly detection algorithm. Specifically, considering the temporal proximity between the gas consumption and the outdoor temperature, the normality detection algorithm adopts a similarity-based deformation correlation to detect normal boiler room users out of abnormal ones. Then, we employ OCSVM as the anomaly detection algorithm to capture various features across multiple data sources, aiming to distinguish gas-theft suspects from the remaining irregular users. Here, the detected normal and abnormal users are fed into the OCSVM for training and prediction, respectively, which can overcome the label scarcity problem. We conduct extensive experiments on a real-world dataset during one heating season. The results demonstrate distinct advantages of our approach over various baselines. We have developed a real-time system on the cloud, providing daily gas-theft suspects for gas companies. [ABSTRACT FROM AUTHOR]

Single-phaseauto-reclosing schemes depend on two main characteristics:(i) secondary arc extinction instant (ii) fault type (temporary or permanent). However, secondary arc quenching time greatly varies from no-arc to prolonged-arc generation depending on the secondary arc current and meteorological factors (wind, thermal buoyancy, etc.). Existing solutions face challenges in making reclosing decisions owing to variable secondary arc quenching instant, non-linear arc voltage characteristics, and varying recovery voltage waveform magnitude. To comprehensively address the discussed problems, an approach based on a particle filter algorithm along with a convolutional neural network is developed. Initially, harmonics of compromised phase are online estimated using a particle filter algorithm, and two indices are developed: (i) estimation of the precise moment of secondary arc extinction (ii) fault nature. Then, extensive simulations are carried out with variable arc resistance, line lengths, no-shunt, and shunt configurations to develop a database based on these two indices. Finally, support vector machine, fully connected deep neural network, and the convolutional neural network are applied to the dataset to predict the reclosing decisions. The comparative study of classifiers confirms the effectiveness of the proposed convolutional neural network in terms of overall model and inter-class accuracy under a noisy and noise-free environment. [ABSTRACT FROM AUTHOR]

Identity authentication based on Doppler radar respiration sensing is gaining attention as it requires neither contact nor line of sight and does not give rise to privacy concerns associated with video imaging. Prior research demonstrating the recognition of individuals has been limited to isolated single-subject scenarios. When two equidistant subjects are present, identification is more challenging due to the interference of respiration motion patterns in the reflected radar signal. In this research, respiratory signature separation techniques are functionally combined with machine learning (ML) classifiers for reliable subject identity authentication. An improved version of the dynamic segmentation algorithm (peak search and triangulation) was proposed, which can extract distinguishable airflow profile-related features (exhale area, inhale area, inhale/exhale speed, and breathing depth) for medium-scale experiments of 20 different participants to examine the feasibility of extraction of an individual’s respiratory features from a combined mixture of motions for subjects. Independent component analysis with the joint approximation of diagonalization of eigenmatrices (ICA-JADE) algorithm was employed to isolate individual respiratory signatures from combined mixtures of breathing patterns. The extracted hyperfeature sets were then evaluated by integrating two different popular ML classifiers, k-nearest neighbor (KNN) and support vector machine (SVM), for subject authentication. Accuracies of 97.5% for two-subject experiments and 98.33% for single-subject experiments were achieved, which supersedes the performance of prior reported methods. The proposed identity authentication approach has several potential applications, including security/surveillance, the Internet-of-Things (IoT) applications, virtual reality, and health monitoring. [ABSTRACT FROM AUTHOR]

Defect pattern detection and classification are challenging for thin-film-transistor liquid-crystal display (TFT-LCD) manufacturing. Limitations of the existing solutions for automatic optical inspection can be traced in part to the lack of a framework within which different existing and new defect patterns can be analyzed, while integrating domain knowledge and effective technologies. This study aims to develop a framework for image-based defect classification that employs the convolution neural networks without using complex and time-consuming image-processing processes in advance. An empirical study was conducted in a leading TFT-LCD manufacturing in Taiwan for validation. The results have shown that the defect patterns can be effectively classified by the proposed convolutional neural networks that outperform the existing approaches such as Support Vector Machine and Random Forest. The developed solution is implemented to effectively support the engineers. [ABSTRACT FROM AUTHOR]

Pedestrian detection is an important but challenging problem in computer vision, especially in human-centric tasks. Over the past decade, significant improvement has been witnessed with the help of handcrafted features and deep features. Here we present a comprehensive survey on recent advances in pedestrian detection. First, we provide a detailed review of single-spectral pedestrian detection that includes handcrafted features based methods and deep features based approaches. For handcrafted features based methods, we present an extensive review of approaches and find that handcrafted features with large freedom degrees in shape and space have better performance. In the case of deep features based approaches, we split them into pure CNN based methods and those employing both handcrafted and CNN based features. We give the statistical analysis and tendency of these methods, where feature enhanced, part-aware, and post-processing methods have attracted main attention. In addition to single-spectral pedestrian detection, we also review multi-spectral pedestrian detection, which provides more robust features for illumination variance. Furthermore, we introduce some related datasets and evaluation metrics, and a deep experimental analysis. We conclude this survey by emphasizing open problems that need to be addressed and highlighting various future directions. Researchers can track an up-to-date list at https://github.com/JialeCao001/PedSurvey. [ABSTRACT FROM AUTHOR]

Series arc fault is one of the main causes of electrical fire. In a three-phase motor with inverter circuit, it is difficult to accurately identify series arc fault occurred in the lines at the back of the inverter, especially when the power supply contains complicated harmonics. To solve this problem, a new feature extraction method of the series arc fault based on the current signal measured at the front of the inverter (CSMFI) was proposed. The series arc fault experiments under different harmonic power supply conditions were carried out in the three-phase motor with inverter circuit. Five-layer decomposition was performed on the CSMFI by using empirical wavelet transform. The attractor track matrix (ATM) of each decomposed signal was established. The fault features of the series arc fault were obtained by calculating and selecting the singular values of the ATM. The series arc fault was identified by using an optimized support vector machine. The effectiveness of the method and its applicability under different experimental conditions were tested. The method can simultaneously identify the series arc fault occurred in the lines at the front or back of the inverter by analyzing the CSMFI. It also has strong anti-interference ability. [ABSTRACT FROM AUTHOR]

An end-to-end supervised learning method is proposed for fault detection in the electric grid using Big Data from multiple Phasor Measurement Units (PMUs). The approach consists of preprocessing steps aimed at reducing data noise and dimensionality, followed by utilization of six classification models considered for detecting faults. Three of the models were variants of Convolutional Neural Network (CNN) architectures that consider a single type of measurement (voltage, current or frequency) at all PMUs or all types together also at all PMUs. CNN based models were compared to traditional methods of Logistic Regression (LR), Multi-layer Perceptron (MLP) and Support Vector Machine (SVM). Evaluation was conducted on two-year data measured by PMUs at 37 locations in a large electric grid. The response variable for classification were extracted from the grid-wide outage event log. Experiments show that CNN-based models outperformed traditional methods on one year out-of-sample outage detection over the entire grid. [ABSTRACT FROM AUTHOR]

The blood pressure (BP) is generally measured using a cuff based sphygmomanometer. However, it is inconvenient to be used. Recently, an alternative solution only using the photoplethesmograms (PPGs) was proposed. In this case, the continuous BP estimation could be performed. First, the features were extracted from the PPGs. Then, a regression network was employed to estimate the BP values. Nevertheless, the accuracy of this approach was not so high. In order to improve the estimation accuracy, this paper proposes to cascade a two layer piecewise neural network to the output of the existing regression network to correct the estimation error. In particular, the overall system is a three layer network. The first layer of the network is the existing regression network. It generates the initial estimated BP values. The second layer of the network consists of the window functions. It segments the range of the BP values into various regions for the further processing. The final layer of the network performs the estimation correction. The performance of our proposed network is evaluated via two practical datasets and three common regression networks including the three layer artificial neural network (ANN) based regression network, the random forest (RF) based regression network and the support vector regression (SVR) based network. For the first dataset, our proposed method with the RF model and the piecewise neural network achieves the systolic BP (SBP) estimation error and the diastolic BP (DBP) estimation error at $3.01{\pm }2.22$ mmHg with the correlation coefficient at 0.926 and $4.43{\pm }3.37$ mmHg with the correlation coefficient at 0.935, respectively. On the other hand, the conventional RF model without the piecewise neural network achieves the SBP estimation error and the DBP estimation error at $5.34{\pm }4.08$ mmHg with the correlation coefficient at 0.740 and $5.89{\pm }4.98$ mmHg with the correlation coefficient at 0.840, respectively. For the second dataset, our proposed method with the RF model and the piecewise neural network achieves the SBP estimation error and the DBP estimation error at $7.91{\pm }8.06$ mmHg with the correlation coefficient at 0.876 and $3.47{\pm }5.59$ mmHg with the correlation coefficient at 0.859, respectively. On the other hand, the conventional RF model without the piecewise neural network achieves the SBP estimation error and the DBP estimation error at $9.77{\pm }9.01$ mmHg with the correlation coefficient at 0.805 and $7.08{\pm }5.55$ mmHg with the correlation coefficient at 0.612, respectively. It can be seen that our proposed network yields the estimated BP values highly correlated to the reference BP values. Also, our proposed method yields the higher accuracies compared to the existing networks. This demonstrates the effectiveness of our proposed network. [ABSTRACT FROM AUTHOR]

During the observation of micromotion targets in space, inverse synthetic aperture radar usually obtains the narrowband and wideband echoes simultaneously. In order to exploit their rich information in target electromagnetic scattering, shape, structure, and motion, this article proposes a recognition method of space micromotion targets based on decision fusion. The proposed method extracts physical features from the radar cross section and the joint time–frequency distribution from narrowband echoes, while extracts the data features from high-resolution range profiles and range-instantaneous-Doppler image by convolution neural network. Finally, particle swarm optimization is adopted to decision-level fusion so as to realize high-precision recognition. The recognition results of electromagnetic simulated data under various conditions have demonstrated the effectiveness and robustness of the proposed method. [ABSTRACT FROM AUTHOR]

Change detection (CD) is among the most important applications in remote sensing (RS) that allows identifying the changes that occurred in a given geographical area across different times. Even though CD systems have seen a lot of progress in RS, their output is either a binary map highlighting the changing area or a semantic change map that indicates the type of change for each pixel. The change maps are often difficult to interpret by end users, and they omit important information such as relationships and attributes of the changed areas. Motivated by the recent advancement of image captioning in the RS community, in this article, we propose to describe the changes over bitemporal images through change sentence descriptions. The aim of this article is to provide a user-friendly interpretation of the occurred changes. To this end, we propose two change captioning (CC) systems that take bitemporal images as input and generate coherent sentence descriptions of the occurred changes. Convolutional neural networks (CNNs) are used to extract discriminative features from the bitemporal images and recurrent neural networks (RNNs) or support vector machines (SVMs) are exploited to generate coherent change descriptions. Furthermore, in the absence of a CC dataset to test our systems, we propose two new datasets. One is based on very high-resolution RGB images, and the other one is based on multispectral RS images. The obtained experimental results show promising capabilities of the proposed systems to generate coherent change descriptions from the bitemporal images. The datasets are available at the following link: https://disi.unitn.it/~melgani/datasets.html. [ABSTRACT FROM AUTHOR]

As an essential reprocessing method, dimensionality reduction (DR) can reduce the data redundancy and improve the performance of hyperspectral image (HSI) classification. A novel unsupervised DR framework with feature interpretability, which integrates both band selection (BS) and spatial noise reduction method, is proposed to extract low-dimensional spectral–spatial features of HSI. We proposed a new neighborhood grouping normalized matched filter (NGNMF) for BS, which can reduce the data dimension while preserving the corresponding spectral information. An enhanced 2-D singular spectrum analysis (E2DSSA) method is also proposed to extract the spatial context and structural information from each selected band, aiming to decrease the intraclass variability and reduce the effect of noise in the spatial domain. The support vector machine (SVM) classifier is used to evaluate the effectiveness of the extracted spectral–spatial low-dimensional features. Experimental results on three publicly available HSI datasets have fully demonstrated the efficacy of the proposed NGNMF-E2DSSA method, which has surpassed a number of state-of-the-art DR methods. [ABSTRACT FROM AUTHOR]

Tool condition monitoring (TCM) is vital to maintain the quality of workpieces during machining. Recently, data-driven methods based on multisensory data have been applied to TCM. The quality of extracted features is a key to realizing a successful data-driven TCM. However, the extracted features in the previous study are focused on the multicollinearity of multisensory data, which is incapable of identifying the informative and discriminative information in the long time period aspect. This article proposed a novel method for TCM using deep spatial-temporal feature extraction and lightweight feature fusion techniques. A key to the proposed method is the extraction of multicollinearity as spatial features (SPs), and the capture of long-range dependencies and nonlinear dynamics as temporal features (TFs), to fully characterize tool wear change using multisensory data. Then, a lightweight feature fusion method is used to fuse SPs, TFs, and statistical features for further removing redundant information employing the kernel-principal component analysis. Finally, support vector machines is used to predict the tool conditions using the fusion feature. Experiments on a milling machine and a gear hobbing machine are carried out to verify the effectiveness and generalization of the proposed method respectively. [ABSTRACT FROM AUTHOR]

Accurate and real-time prediction of network traffic can not only help system operators allocate resources rationally according to their actual business needs but also help them assess the performance of a network and analyze its health status. In recent years, neural networks have been proved suitable to predict time series data, represented by the model of a long short-term memory (LSTM) neural network and a temporal convolutional network (TCN). This article proposes a novel hybrid prediction method named SG and TCN-based LSTM (ST-LSTM) for such network traffic prediction, which synergistically combines the power of the Savitzky–Golay (SG) filter, the TCN, as well as the LSTM. ST-LSTM employs a three-phase end-to-end methodology serving time series prediction. It first eliminates noise in raw data using the SG filter, then extracts short-term features from sequences applying the TCN, and then captures the long-term dependence in the data exploiting the LSTM. Experimental results over real-world datasets demonstrate that the proposed ST-LSTM outperforms state-of-the-art algorithms in terms of prediction accuracy. Note to Practitioners—This work considers real-time and high-accuracy prediction of network traffic. It is highly important to well predict network traffic by capturing long-term dependence and effectively extracting high- and low-frequency information from time series data. Yet, it is a big challenge to achieve it because there are unstable characteristics and strong nonlinear features in the network traffic due to continuous expansion of network scale and fast emergence of new services. Current prediction methods usually have oversimplified theoretical assumptions, need significant time and memory, or suffer problems of gradient disappearance or early convergence. Thus, they fail to effectively capture the nonlinear characteristics of large-scale network sequences. This work proposes a hybrid prediction method named SG and TCN-based LSTM (ST-LSTM), which integrates the merits of the Savitzky–Golay filter, the temporal convolutional network (TCN), and the long short-term memory (LSTM), serving as smoothing time series, capturing short-term local features, and capturing long-term dependence, respectively. Experimental results based on the real-life dataset demonstrate that it achieves better prediction accuracy than its state-of-the-art peers, including the TCN and the LSTM. It can be readily implemented and deployed in many real-life industrial areas including smart city, edge computing, cloud computing, and data centers. [ABSTRACT FROM AUTHOR]

Network alignment is a critical task in a wide variety of fields. Many existing works leverage on representation learning to accomplish this task without eliminating domain representation bias induced by domain-dependent features, which yield inferior alignment performance. This paper proposes a unified deep architecture (DANA) to obtain a domain-invariant representation for network alignment via an adversarial domain classifier. Specifically, we employ the graph convolutional networks to perform network embedding under the domain adversarial principle, given a small set of observed anchors. Then, the semi-supervised learning framework is optimized by maximizing a posterior probability distribution of observed anchors and the loss of a domain classifier simultaneously. We also develop a few variants of our model, such as, direction-aware network alignment, weight-sharing for directed networks and simplification of parameter space. Experiments on three real-world social network datasets demonstrate that our proposed approaches achieve state-of-the-art alignment results. [ABSTRACT FROM AUTHOR]

Vision-based fire detection systems have been significantly improved by deep models; however, higher numbers of false alarms and a slow inference speed still hinder their practical applicability in real-world scenarios. For a balanced trade-off between computational cost and accuracy, we introduce dual fire attention network (DFAN) to achieve effective yet efficient fire detection. The first attention mechanism highlights the most important channels from the features of an existing backbone model, yielding significantly emphasized feature maps. Then, a modified spatial attention mechanism is employed to capture spatial details and enhance the discrimination potential of fire and non-fire objects. We further optimize the DFAN for real-world applications by discarding a significant number of extra parameters using a meta-heuristic approach, which yields around 50% higher FPS values. Finally, we contribute a medium-scale challenging fire classification dataset by considering extremely diverse, highly similar fire/non-fire images and imbalanced classes, among many other complexities. The proposed dataset advances the traditional fire detection datasets by considering multiple classes to answer the following question: what is on fire? We perform experiments on four widely used fire detection datasets, and the DFAN provides the best results compared to 21 state-of-the-art methods. Consequently, our research provides a baseline for fire detection over edge devices with higher accuracy and better FPS values, and the proposed dataset extension provides indoor fire classes and a greater number of outdoor fire classes; these contributions can be used in significant future research. Our codes and dataset will be publicly available at https://github.com/tanveer-hussain/DFAN. [ABSTRACT FROM AUTHOR]

Considering the multiple observations from satellites with different types of sensors, the available object data are regarded as different combinations of multisource heterogeneous data. To achieve accurate object recognition, the dimension reduction (DR) technology is important to capture the low-dimensional discriminative representation containing complementary information from multisource data, while off-the-shelf DR methods can only handle input represented as vector or homogeneous tensors and fail to deal with multisource data represented as heterogeneous tensors. In addition, the existing DR methods are classifier-independent, making it difficult to ensure effective recognition under a specific classifier. To solve these problems, the DR method for Heterogeneous Tensors based on Graph-based and Classifier-oriented Embedding (HTGCE) method is proposed to learn the low-dimensional representation of multisource data using different combinations of samples. First, self-learning adjacency matrices are constructed to capture the local structure of different combinations of multisource data autonomously. Then, unlike the classifier-independent discriminant term used in existing DR methods, a classifier-oriented discriminant term is constructed to enhance the specific classifier-based recognition results. Furthermore, the reconstruction error minimization term is created to enable DR results to inherit the main information of the original data. Moreover, an adaptive weight factor is built to balance the importance of different sources for object recognition. Finally, an alternative optimization strategy is presented to solve the optimization problem of HTGCE. Using the multiresolution multiangle optical dataset and the paired optical and SAR dataset, the experimental results demonstrate that the HTGCE outperforms typical vector- and tensor-based DR methods in terms of recognition accuracy. [ABSTRACT FROM AUTHOR]

Remote sensing hyperspectral images are very useful for land cover classification because of their rich spatial and spectral information. However, hyperspectral image acquisition and pixel labeling are laborious and time-consuming, so few-shot learning methods are considered to solve this problem. Deep learning has gradually been used for few-shot hyperspectral classification, but there are some problems. The feature extraction network based on deep learning requires too many parameters to be trained, resulting in a huge network model, which is not conducive to deployment on remote sensing data acquisition equipment. Moreover, due to the lack of label samples, the algorithm based on deep learning is more prone to overfitting. To solve the above problems, considering the advanced characteristics of the kernel method in dealing with nonlinear, small sample, and high-dimensional data, we propose a small scale high precision network called 3-D convolution random Fourier features (3-DCRFF) based on the random Fourier feature (RFF) kernel approximation, which is the 3-DCRFF network. First, we combine 3-D convolution with RFF as the basic structure of the network to extract the spatial and spectral features of HSI cubes. Second, we use a classifier based on attention mechanism to classify feature vectors to obtain recognition probability. Finally, the network parameters are solved from the perspective of Bayesian optimization, and the synthetic gradient optimization method is designed and implemented to realize the fast learning of the network. A large number of experiments HSI classification experiments were performed on University of Pavia (UP), Pavia Center (PC), Indian Pines (IP), and Salinas standard remote sensing datasets, the results show that our algorithm outperforms most state-of-the-art algorithms on few-shot classification. [ABSTRACT FROM AUTHOR]

One of the main challenges in multi-source domain adaptation is how to reduce the domain discrepancy between each source domain and a target domain, and then evaluate the domain relevance to determine how much knowledge should be transferred from different source domains to the target domain. However, most prior approaches barely consider both discrepancies and relevance among domains. In this paper, we propose an algorithm, called Iterative Refinement based on Feature Selection and the Wasserstein distance (IRFSW), to solve semi-supervised domain adaptation with multiple sources. Specifically, IRFSW aims to explore both the discrepancies and relevance among domains in an iterative learning procedure, which gradually refines the learning performance until the algorithm stops. In each iteration, for each source domain and the target domain, we develop a sparse model to select features in which the domain discrepancy and training loss are reduced simultaneously. Then a classifier is constructed with the selected features of the source and labeled target data. After that, we exploit optimal transport over the selected features to calculate the transferred weights. The weight values are taken as the ensemble weights to combine the learned classifiers to control the amount of knowledge transferred from source domains to the target domain. Experimental results validate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

The transient overvoltages in offshore wind farms caused by faults or frequent operations of electrical equipment are particularly severe. In order to classify the internal transient overvoltages in offshore wind farms, this research firstly proposes a feature extraction method based on sparse decomposition using alternating direction method of multipliers, which extracts the harmonic and impulsive components from transient overvoltages. Then based on the impulsive components and transient overvoltages, two identification features are constructed. One is an impulsive feature and the other one is a ratio feature between impulsive energy values and original energy values. Finally, based on the constructed features, a support vector machine is employed to identify different types of internal transient overvoltages. Extensive simulations and experiments are performed to verify the effectiveness of the proposed feature extraction method, which is also compared with the widely used conventional wavelet algorithms and the mathematical morphology. Results show the proposed feature extraction method is capable of classifying and discriminating among various types of internal transient overvoltages with a strong self-adaptive ability and an improved accuracy. [ABSTRACT FROM AUTHOR]

In the controller development process, it is required to verify whether control functions operate normally, without fault or anomaly. Many control functions with complex system structures require various actual tests and test data analysis for verification. This paper presents an anomaly detection algorithm to verify control functions and to more efficiently analyze test data of a hybrid control unit (HCU) of a hybrid electric vehicle (HEV). The anomaly detection algorithm automatically detects anomaly of control functions from the test data, instead of manual labor by engineers. The examined target HCU control functions in this paper are the engine clutch engagement/disengagement control function and the engine start cooperative control function, which are crucial functions of transmission mounted electric drive (TMED) HEVs. A data-driven approach using a one-class support vector machine (SVM) is used to make it easy to apply various control functions. Actual vehicle test data is examined by the algorithm to verify the control functions at an actual vehicle level. The developed anomaly detection algorithm demonstrates feasibility and effectiveness of the proposed algorithm in detecting not only prior-known anomalies but also prior-unknown anomalies. Because other vehicle control functions have similar characteristics with the target functions of this paper, this algorithm is expected to be successfully applied to other HCU control functions. [ABSTRACT FROM AUTHOR]

Railway point machines (RPMs) is one of the most vital devices closely related to the efficiency and safety of train operation. Considering the advantages of contactlessness and easy-to-collect of sound signals, a novel sound-based fault diagnosis method for RPMs is proposed. First, fractional calculus is introduced to wavelet packet decomposition energy entropy (WPDE). Fractional WPDE (FWPDE) is then proposed, which is verified to be a more effective tool for fault feature representation. Second, coarse-grain process is firstly introduced to FWPDE. Novel feature named multi-scale FWPDE is developed, which can significantly improve fault diagnosis accuracy. Third, to select optimal feature set and optimize the hyperparameters of support vector machine (SVM) at the same time, a synchronous optimization strategy based on binary particle swarm optimization (BPSO) is presented, which can further improve the diagnosis accuracy. The superiority and effectiveness of the proposed method are verified by comparing to some existing fault diagnosis methods. The diagnosis accuracies of reverse-normal and normal-reverse switching processes reach 99.33% and 99.67%, respectively. Especially, the proposed method is suitable for diagnosis of similar faults, which can also provide reference for similar research fields. [ABSTRACT FROM AUTHOR]

With the continuous progress of computer deep learning technology, convolutional neural network (CNN), as a representative approach, provides a unique solution for hyperspectral image (HSI) classification. However, the parameters of CNN cannot be well-tuned when the number of training samples is insufficient, resulting in unsatisfactory classification performance. To tackle the thorny problem, a deep ensemble CNN method based on sample expansion for HSI classification is studied in this article. In particular, spatial information is first extracted and fused with original spectral bands to help classifiers obtain discriminant spectral–spatial features. Then, we use the pixel-pair feature (PPF) to expand the number of training samples so that the parameters of CNN structure can be fully trained. In addition, deep ensemble CNN is employed in this article, enabling the trained model to obtain better generalization ability and more robust classification results. Ultimately, the proposed method is applied to classify four widely used hyperspectral datasets. Experimental results show that the studied approach yields higher classification accuracy than some CNN-based methods even under the condition of small-size training set. [ABSTRACT FROM AUTHOR]

In hyperspectral image (HSI) classification, using spatial information as a supplement to spectral information is an effective way to improve classification accuracy. In this article, a novel robust complementary method using spectral-spatial information is proposed to reduce the information loss in feature extraction, thus improving the classification effect. In short, two complementary feature extraction stages are used to get the probability maps for decision fusion. In the stage of pre-processing feature extraction, we propose an adaptive cubic total variational smoothing method (ACTVSP), which is first proposed and applied in the remote sensing research field, to obtain the first-stage probability map. At the same time, we utilize edge-preserving filtering in the post-processing stage and obtain the second probability map by means of pixel-level classifier. Finally, the probability-like maps obtained in the above two stages are integrated by decision fusion rules. Experiments on ten public datasets show the effectiveness of our proposed method and demonstrate the superiority of distinguishing different land covers on the basis of very few training samples. Therefore, it can be applied to practical applications in different scenes. [ABSTRACT FROM AUTHOR]

Improving ship classification performance in synthetic aperture radar (SAR) imagery by the methods based on transfer learning (TL) is a newly emerging research topic and has great potential. The existing studies merely address the problem of TL from a single source domain (SD) [automatic identification system (AIS) or optical remote sensing (ORS)] to the target domain (TD) (SAR) based on the homogeneous TL (HoTL), which requires all domains to be represented by homogeneous features with the same dimensions. Our work takes a step forward and attempts to address a more meaningful and challenging problem that transfers knowledge from multiple SDs for the purpose of exploring and exploiting complementarity cross SDs to assist ship classification in the TD. To this end, our study develops a multisource heterogeneous transfer learning (MS-HeTL) method that liberates the restriction of utilizing the exact same features for all domains, allows each domain represented by a more appropriate feature, and, thus, improves the ship classification performance even further. Specifically, we first propose multisource heterogeneous feature augmentation (MS-HFA) to effectively solve the challenges brought by feature heterogeneity and fully exploit the complementarity cross domains. Then, a support vector machine (SVM) classification framework is specific-designed to be incorporated with the augmented feature representations in the common space to conduct knowledge transfer cross domains. Extensive experiments on two benchmark datasets, HR-SAR and FUSAR, show that the proposed method outperforms the existing methods and demonstrates its effectiveness and advantages. All datasets and source codes are available at https://github.com/BUCT-RS-ML/MS-HeTL-via-MS-HFA. [ABSTRACT FROM AUTHOR]

Studies on the classification of hyperspectral images (HSIs) based on deep learning are in full swing, especially the spectral–spatial dependent global learning (SSDGL) framework, which is both efficient and robust. However, the global convolutional long short-term memory (GCL) module under this framework fails to take full consideration of the spectral characteristics contained in HSIs, and the hierarchically balanced (H-B) sampling strategy introduced in this framework prevents the training process from converging smoothly. In this article, we develop a novel regularized spectral–spatial global learning (RSSGL) framework. Compared with SSDGL, the proposed framework mainly makes three improvements. Above all, aiming at the problem that the GCL module used in SSDGL cannot fully tap the local spectral dependence, we apply 3-D convolution to the gated units of long short-term memory (LSTM) as an alternative to the GCL module for adjacent and nonadjacent spectral dependencies learning. Furthermore, to extract the most discriminative features, an improved statistical loss regularization term is developed, in which we introduce a simple but effective diversity-promoting condition to make it more reasonable and suitable for deep metric learning in HSI classification. Finally, to effectively address the performance oscillation caused by the H-B sampling strategy, the proposed framework adopts an early stopping strategy to save and restore the optimal model parameters, making it more flexible and stable. Experiments conducted on three representative datasets show that the proposed RSSGL has superior classification performance compared with the existing relatively excellent research methods. The source code is released at https://github.com/swiftest/RSSGL. [ABSTRACT FROM AUTHOR]

This article introduces a multiclass classification approach accustoming the benefits of partitioning-based strategies and hierarchical techniques. The proposed hierarchical framework creates a hierarchy with the notion of grouping classes with similar traits as one group. It overcomes the deficiency of the existing multiclass extension approaches, viz., nonlinearity, imbalanced class classification, and increasing classification cost with increasing number of classes. The hierarchical framework presents the idea of decomposing several classes hierarchically, where every cluster contains a set of classes having similar traits. The approach aims to maximize the intercluster distance and minimize the intracluster distribution. The effectiveness of the proposed method is evaluated on real-world and complex problems of plant recognition. Three leaf image datasets are considered for performance evaluation using a support vector machine. The results signify that the proposed approach for multiclass classification is an efficient approach with significantly improved recognition accuracy. It is a robust and effective approach with the least computational cost. The speedup factor of the proposed approach in the binary structure is 16, 6.5, and 5.5 as compared to a one-versus-one traditional support vector machine for Flavia, Swedish, and self-collected leaf datasets, respectively. [ABSTRACT FROM AUTHOR]

Travel mode classification within travel survey data sets, especially light-duty vehicle (LDV) trips, is foundational, though nontrivial, to emerging mobility systems, travel behavior analysis, and fuel consumption estimation. Current travel mode detection approaches require well-sampled and balanced data sets with ground-truth travel mode labels. The detection approaches are rarely applied and validated on large-scale, real-world data sets, which may not satisfy the data set requirements. This article proposes an LDV trip detection model as a supplement to current travel mode detection methods for the case when the training set is highly (and/or completely) unbalanced, to the extent that classical machine learning approaches become difficult or impossible to deploy. The proposed model uses a novelty detection technique—one-class support vector machines—and a novel exhaustive feature extraction technique on continuous time series data [i.e., global positioning system (GPS) speed profiles] for single-mode trip trajectories. Training and validation of the model are conducted on a large-scale, real-world data set. The proposed method accurately identifies LDV trips from a broad set of multimodal trips by leveraging a wealth of pre-existing in-vehicle GPS travel data. Additional sensitivity analysis sheds light on the optimal training size and feature selection, which will benefit applications limited by highly imbalanced data. This article also discusses performance comparison with regular machine learning approaches, the model's robustness, and the potential to extend the proposed model to multimodal trip prediction. [ABSTRACT FROM AUTHOR]

The wire bonding process is one of the most critical processes in semiconductor packaging. The electrical performance and reliability of IC chips must be evaluated by probe testing before wire bonding. In the probe test, the probe usually leaves traces on the surface of the pad. Large probe marks tend to cause a decrease in bond adhesion and so increase the possibility of ball bond lifting from the pads. Shear force is an important measure of the adhesion between the ball and pad. The prediction of the shear force helps to understand the bonding quality in advance. In this study, the six features of the probe marks were extracted by the automatic image recognition method. Using three machine learning techniques (logistic regression, support vector machine, and random forest) based on principal component analysis (PCA), the shear force is estimated based on six features before wire bonding. The results indicate that the proposed PCA-based random forest could identify bad chips with 97.92% accuracy before wire bonding. Unnecessary chips can be found early, and unnecessary wire bonding can be avoided, thereby saving process time and cost. The proposed method would improve the efficiency and quality yield of the semiconductor packaging process. [ABSTRACT FROM AUTHOR]

Semiconductors are essential components in many electronic devices. Because wafers are produced quickly and in large quantities, defects occur that adversely affect semiconductor properties. This makes it necessary to install powerful and robust inspection systems which use artificial intelligence techniques in the early stages of the manufacturing chain in order to detect and classify those defects. This paper proposes a method for defect detection and classification on images of semiconductor wafer materials obtained by means of a scanning electron microscope based in the following stages: (i) use of computer vision techniques to isolate the defect from the background; (ii) use of several descriptors based on shape, size, texture, histogram, and key-points to create a feature vector for the characterization of the defect; (iii) application of an exhaustive search as a feature selection method to determine the optimal subset of feature descriptors; and (iv) evaluation of the feature descriptors by using a support vector machine classifier providing the optimal set with highest F1-score metrics. Finally, the effectiveness of the proposed approach is compared with five popular feature selection methods, reporting better classification results than the latter. [ABSTRACT FROM AUTHOR]

Feature selection aims at choosing a subset of features to represent the original feature space. In practice, however, it is hard to achieve desirable performance due to limited training data. To alleviate this issue, we propose a novel problem named feature selection with multi-source transfer where the privileged information from another data source or modality– only available during the training phase, is exploited to improve the performance of feature selection. To be exact, we propose a novel objective function that formulates the privileged information into feature selection. Moreover, an efficient optimization algorithm is introduced to solve the proposed problem of high dimension. Extensive experimental results demonstrate that the proposed algorithm significantly outperforms several popular algorithms, especially when the training data size and the selected feature size are small. [ABSTRACT FROM AUTHOR]

Objective: Myoelectric control requires fast and stable identification of a movement from data recorded from a comfortable and straightforward system. Methods: We consider a new real-time pre-processing method applied to a single differential surface electromyogram (EMG): deconvolution, providing an estimation of the cumulative firings of motor units. A 2 channel-10 class finger movement problem has been investigated on 10 healthy subjects. We have compared raw EMG and deconvolution signals, as sources of information for two specific classifiers (based on either Support Vector Machines or k-Nearest Neighbours), with classical time-domain input features selected using Mutual Component Analysis. Results: Using the proposed pre-processing technique, classification performances statistically improve. For example, the true positive rates of the best-tested configurations were 80.9% and 86.3% when using the EMG and its deconvoluted signal, respectively. Conclusion: Even considering the limited dataset and range of classification approaches investigated, our preliminary results indicate the potential usefulness of the deconvolution pre-processing. Significance: Deconvolution of EMG is a fast pre-processing that could be easily embedded in different myoelectric control applications. [ABSTRACT FROM AUTHOR]

Objective: Dexterous hand motion is critical for object manipulation. Electrophysiological studies of the hand are key to understanding its underlying mechanisms. High-density electromyography (HD-EMG) provides spatio-temporal information about the underlying electrical activity of muscles, which can be used in neurophysiological research, rehabilitation and control applications. However, existing EMG electrodes platforms are not muscle-specific, which makes the assessment of intrinsic hand muscles difficult. Methods: Muscle-specific flexible HD-EMG electrode arrays were developed to capture intrinsic hand muscle myoelectric activity during manipulation tasks. The arrays consist of 60 individual electrodes targeting 10 intrinsic hand muscles. Myoelectric activity was displayed as spatio-temporal amplitude maps to visualize muscle activation. Time-domain and temporal-spatial HD-EMG features were extracted to train cubic support vector machine machine-learning classifiers to classify the intended user motion. Results: Experimental data was collected from 5 subjects performing a range of 10 common hand motions. Spatio-temporal EMG maps showed distinct activation areas correlated to the muscles recruited during each movement. The thenar muscle fiber conduction velocity (CV) was estimated to be at 4.7±0.3 m/s for all subjects. Hand motions were successfully classified and average accuracy for all subjects was directly related to spatial resolution based on the number of channels used as inputs; ranging from 74±4% when using only 5 channels and up to 92±2% when using 41 channels. Temporal-spatial features were shown to provide increased motion-specific accuracy when similar muscles were recruited for different gestures. Conclusions: Muscle-specific electrodes were capable of accurately recording HD-EMG signals from intrinsic hand muscles and accurately predicting motion. Significance: The muscle-specific electrode arrays could improve electrophysiological research studies using EMG decomposition techniques to assess motor unit activity and in applications involving the analysis of dexterous hand motions. [ABSTRACT FROM AUTHOR]

Image scene classification aiming to assign specific semantic labels for each image is vitally important for the applications of remote sensing (RS) data. In real world, since the observation environment is open and dynamic, RS images are collected sequentially and the numbers of images and classes grow rapidly over time. Most existing scene classification methods are offline learning algorithms, which are inefficient and unscalable for this scenario. In this article, an incremental learning with open-set recognition (ILOSR) framework is proposed for RS image scene classification in the open and dynamic environment, which can identify the unknown classes from a stream of data and learn these new classes incrementally. Specifically, a controllable convex hull-based exemplar selection strategy is designed to address the catastrophic forgetting issue in incremental learning, which can reduce training time and memory footprint effectively. In addition, a new loss function based on prototype learning and uncertainty measurement is proposed for OSR to enhance the interclass discrimination and intraclass compactness of the learned deep features. Experimental results on real RS datasets demonstrate that the proposed method can not only outperform the state-of-the-art approaches on offline classification, incremental learning, and OSR problem separately but also achieve better and more stable performance in the experiments for ILOSR. [ABSTRACT FROM AUTHOR]

Feature selection is of great importance to make prediction for process variables in industrial production. An embedded feature selection method, based on relevance vector machines with an approximated marginal likelihood function, is proposed in this study. By setting hierarchical prior distributions over the model weights and the parameters of the automatic relevance determination kernel function, respectively, a Gaussian approximation method is designed to approximate the intractable exact marginal likelihood by using the law of the total expectation and the total covariance. Furthermore, in this study, the joint posterior distribution over the model weights and the kernel parameters is estimated by combining the Gibbs sampling with a Laplace approximation. Thus, feature selection is performed by examining the posterior over the kernel parameters. To verify the performance of the proposed method, a series of benchmark datasets and two practical industrial datasets are employed. The experimental results demonstrate that the proposed method not only produces higher prediction accuracy than other methods but also performs better in feature selection, especially in industrial case. [ABSTRACT FROM AUTHOR]

The hyperspectral images (HSIs) classification technique has received widespread attention in the field of remote sensing. However, how to achieve satisfactory classification performance in the presence of a large amount of noise is still a problem worthy of consideration. In this article, a local correntropy matrix (LCEM)-based spatial–spectral feature representation method is proposed for HSI classification. Motivated by the successful application of information-theoretic learning (ITL), we propose to adopt correntropy matrix to represent the spatial–spectral features of HSI. Specifically, the dimension reduction is first performed on the original hyperspectral data. Then, for each pixel, we select its local neighbors within a sliding window using cosine distance for the construction of the LCEM. In this way, each pixel can be characterized as an LCEM. Finally, all the correntropy matrices are fed into a support vector machine (SVM) for final classification. In addition, we also propose a novel way to determine the size of the local window based on standard deviation. Because the LCEM as the feature descriptor can characterize discriminative spatial–spectral features, the proposed method has shown great interclass separability and intraclass compactness. Compared with other advanced approaches, the proposed LCEM method has achieved competitive performance in both evaluation indexes and visual effects, especially when the training size is very small. [ABSTRACT FROM AUTHOR]

A large amount of synchrophasor data in the wide area measurement system (WAMS) needs to be collected and transmitted to the phasor data concentrator, thereby increasing the possibility of being attacked by hackers. The attacked data are therefore hidden into the normal synchrophasor data so that the synchrophasor data based application will be affected. To remedy this problem, an identification framework is proposed to detect the data cyber-attack in WAMS utilizing variational mode decomposition (VMD) and multifusion support vector machine (MSVM). First, VMD is used to transform the attacked data into multiple modal components. Thereafter, a novel MSVM is employed to classify the deterministic features using the proposed linear combined multikernel (LCM). This LCM can fuse multiple types of features, including the time, frequency, and statistical domains of the synchrophasor data. Utilizing the actual data from FNET/GridEye, different experiments are conducted under multiple attack strengths and types. The results demonstrate that the identification framework has higher precision and robustness compared with other conventional classifiers. [ABSTRACT FROM AUTHOR]

Fused magnesium furnace (FMF) is the vital equipment for magnesia refractory production. The melting process of FMF is subject to melting temperature, magnesite quality and composition, and much more dynamic factors, which are prone to abnormal conditions such as overheating, abnormal exhausting, semi-molten, and affect the stability and safety of the production seriously. Due to the difficulties of effectively monitoring and accurately identifying the abnormal conditions with the existing methods, this article proposes a novel method for identifying the abnormal conditions of FMF based on deep learning and multi-information fusion. First, the Wasserstein distance based generative adversarial networks with gradient penalty is developed to solve the problem of obtaining overheating condition image samples with obvious visual characteristics. Then, the image-based deep convolutional neural network model is established to recognize the abnormal condition images and extract image features whose training set is the combination of the generated images and the original images. Finally, based on multisource information fusion, the traditional process current data are fused with the extracted image features, and a support vector machine model is established to complete the classification and identification of the FMF working conditions. Various experiments on actual industrial data show that the proposed method is effective and advanced. [ABSTRACT FROM AUTHOR]

In this paper, we address challenges of detecting instances from emerging classes over a non-stationary data stream during data classification. In particular, data instances from an entirely unknown class may appear in a data stream over time. Existing classification techniques utilize unsupervised clustering to identify emergence of such data instances. Unfortunately, they make strong assumptions which are typically invalid in practice; (i) Most instances associated with a class are closer to each other in feature space than instances associated with different classes, (ii) Covariates of data are normalized through an oracle to overcome the effect of a few data instances having large feature values, and (iii) Labels of instances from emerging classes are readily available soon after detection. To address the challenges that occur in practice when the above assumptions are weak, i.e., instances of each class are scattered and the true labels of novel class instances are sparsely available, we propose a practical semi-supervised emerging class detection framework. Particularly, we aim to identify similar data instances within local regions in feature space by incorporating a mutual graph clustering mechanism. We also perform online normalization along the data stream instead of assuming an oracle, and propose a classification technique that uses only a small amount of true labels for training and emerging class detection. Our empirical evaluation of this framework on real-world datasets demonstrates its superiority of classification performance compared to existing methods, while using significantly fewer labeled instances. [ABSTRACT FROM AUTHOR]

To implement the human–robot interactions in a noisy environment, the speech improvement-based (SIB) stratified adaptive finite-time saturation control (SAFTSC) for omnidirectional service robot (OSR) is developed. From the outset, the feature vectors of nine designed speech commands are extracted from their frequency signals and then trained by multiclass support vector machine. Two background noises are on-line filtered with the characteristic: “the smaller error in power spectrum is, the larger recovery from noisy power spectrum is.” Comparisons among without or with noise, and filtering are addressed. To achieve the zero pose error of OSR in finite time, an adaptive finite-time indirect trajectory (AFTIT) is constructed. To track the AFTIT with the zero error in finite time, the adaptive finite-time saturation control (AFTSC) is also established. Both AFTIT and AFTSC possess nonlinear switching gain increasing the high-frequency motion capability to fulfill the classified speech command. Simply put, the proposed SIB stratified AFTSC includes the speech improvement for classification, the AFTIT, and the AFTSC. Besides the stability of the closed-loop system is verified by the Lyapunov stability theory, three categories of SIB experiments are compared. [ABSTRACT FROM AUTHOR]

Recently, blockchain technology has become a topic in the spotlight but also a hotbed of various cybercrimes. Among them, phishing scams on blockchain have been found to make a notable amount of money, thus emerging as a serious threat to the trading security of the blockchain ecosystem. In order to create a favorable environment for investment, an effective method for detecting phishing scams is urgently needed in the blockchain ecosystem. To this end, this article proposes an approach to detect phishing scams on Ethereum by mining its transaction records. Specifically, we first crawl the labeled phishing addresses from two authorized websites and reconstruct the transaction network according to the collected transaction records. Then, by taking the transaction amount and timestamp into consideration, we propose a novel network embedding algorithm called trans2vec to extract the features of the addresses for subsequent phishing identification. Finally, we adopt the one-class support vector machine (SVM) to classify the nodes into normal and phishing ones. Experimental results demonstrate that the phishing detection method works effectively on Ethereum, and indicate the efficacy of trans2vec over existing state-of-the-art algorithms on feature extraction for transaction networks. This work is the first investigation on phishing detection on Ethereum via network embedding and provides insights into how features of large-scale transaction networks can be embedded. [ABSTRACT FROM AUTHOR]

Wafer bin maps contain vital information that helps semiconductor manufacturers to identify the root causes and defect pattern failures in wafers. Conventional manual inspection techniques in inspecting these failures are labour intensive and cause prolonged production cycle time. Therefore, automatic inspection techniques can solve this problem. This paper proposes a deep learning approach based on deep convolutional generative adversarial network (DCGAN) and a new Capsule Network (WaferCaps). DCGAN was used to upsample the original dataset and therefore increase the data used for training and balance the classes at the same time. While WaferCaps was proposed to classify the defect patterns according to eight classes. The performance of our proposed DCGAN and WaferCaps was compared with different deep learning models such as the original Capsule Network (CapsNet), CNN, and MLP. In all of our experiment, WM-811K dataset was used for the data upsampling and training. The proposed approach has shown an effective performance in generating new synthetic data and classify them with training accuracy of 99.59%, validation accuracy of 97.53% and test accuracy of 91.4%. [ABSTRACT FROM AUTHOR]

In medical imaging, quantitative measurements have shown promise in identifying diseases by classifying normal versus pathological parameters from tissues. The support vector machine (SVM) has shown promise as a supervised classification algorithm and has been widely used. However, the classification results typically identify a category of abnormal tissues but do not necessarily differentiate progressive stages of a disease. Moreover, the classification result is typically provided independently as a supplement to medical images, which contributes to an overload of information sources in the clinic. Hence, we propose a new imaging method utilizing the SVM to integrate classification results into medical images. This framework is called disease-specific imaging (DSI) that produces a color overlaid highlight on B-mode ultrasound images indicating the type, location, and severity of pathology from different conditions. In this article, the SVM training was performed to construct hyperplanes that can differentiate normal, fibrosis, steatosis, and pancreatic ductal adenocarcinoma (PDAC) metastases in livers based on ultrasound echoes. Also, cluster centroids for specific diseases define unique disease axes, and the inner product between measured features and any disease axis selected by the SVM quantifies the disease progression. The features were measured from 2794 ultrasound frames using the H-scan analysis, attenuation estimation, and B-mode image analysis. The performance of our proposed DSI method was evaluated for a preclinical model of steatosis (${n} =400$ frames). The contribution of each feature was assessed, and the results were compared with ground truth from histology. Moreover, the images generated by our DSI were compared with earlier imaging methods of B-mode, H-scan, and histology. The comparisons demonstrate that DSI images yield higher sensitivity to monitor progressive steatosis than B-mode and H-scan and provide a comparable performance with the histology. For the parameter comparison, DSI and H-scan resulted in similar correlation with histology (${r}_{s} =0.83$) but higher than attenuation (${r}_{s} =0.73$) and B-mode (${r}_{s} =0.47$). Therefore, we conclude that DSI utilizing the SVM applied to steatosis can visually represent the classification results with color highlighting, which can simplify the interpretation of classification compared to the traditional SVM result. We expect that the proposed DSI can be used for any medical imaging modality that can estimate multiple quantitative parameters at high resolution. [ABSTRACT FROM AUTHOR]

With the increasing development of renewable energy and high-voltage direct current (HVDC) transmission technology, the integration of power electronics-interfaced components has caused sub-/super-synchronous inter-harmonics into power signals, yielding the subsynchronous oscillations (SSOs). Online monitoring and identification of SSOs are crucial to the secure and stable operation of power systems. This paper proposes a data driven SSOs identification method using realistic phasor measurement unit (PMU) data. Three features are extracted via the analysis of a large number of historical PMU data, which can reflect the SSO features and the interference level. An SSOs identification method is proposed through a multiple-support vector machine (SVM) model. It allows us to adaptively select the proper classifier among the multiple-SVM models according to the interference level estimated by the extracted features. This significantly improves the identification accuracy under both low and high interference inputs of a practical power grid. An automatic update function based on incremental learning is also developed to enhance the model performance when the SSO data are subject to new features. Experimental results using field-measured PMU data validate the advantages of the proposed method over other alternatives. [ABSTRACT FROM AUTHOR]