Feature Extraction Tasks Research Articles

DCNN for condition monitoring and fault detection in rotating machines and its contribution to the understanding of machine nature

Rotating machines are critical equipment in many processes, and failures in their operation can have serious implications. Consequently, fault detection in rotating machines has been widely investigated. Conventional detection systems include two blocks: feature extraction and classification. These systems are based on manually engineered features (ball pass frequencies, RMS value, kurtosis, crest factor, etc.) and therefore require a high level of human expertise (it is a human who designs and selects the most appropriate set of features to perform the classification). Instead, we propose a system for condition monitoring and fault detection in rotating machines based on a 1-D deep convolutional neural network (1D DCNN), which merges the tasks of feature extraction and classification into a single learning body. The proposed system has been designed for use on a rotating machine with seven possible operating states and it proves to be able to determine the operating condition of the machine almost as accurately as conventional feature-engineered classifiers, but without the need for prior knowledge of the machine. The proposed system has also reported good classification on a bearing fault dataset from another machine, thus demonstrating its capability to monitor the condition of different machines. Finally, the analysis of the features learned by the deep model has revealed valuable and previously unknown machine information, such as the rotational speed of the machine or the number of balls in the bearings. In this way, our results illustrate not only the good performance of CNNs, but also their versatility and the valuable information they could provide about the monitored machine.

Robust Image Feature Extraction via Approximate Orthogonal Low-Rank Embedding

Feature extraction (FE) plays an important role in machine learning. In order to handle the “dimensionality disaster”problem, the usual approach is to transform the original sample into a low-dimensional target space, in which the FE task is performed. However, the data in reality will always be corrupted by various noises or interfered by outliers, making the task of feature extraction extremely challenging. Thence, we propose a novel image FE method via approximate orthogonal low-rank embedding (AOLRE), which adopts an orthogonal matrix to reserve the major energy of the samples, and the introduction of the ℓ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> , <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> -norm makes the features more compact, differentiated and interpretable. In addition, the weighted Schatten p-norm is adopted in this model for fully exploring the effects of different ranks while approaching the original hypothesis of low rank. Meanwhile, as a robust measure, the correntropy is applied in AOLRE. This can effectively suppress the adverse influences of contaminated data and enhance the robustness of the algorithm. Finally, the introduction of the classification loss item allows our model to effectively fit the supervised scene. Five common datasets are used to evaluate the performance of AOLRE. The results show that the recognition accuracy and robustness of AOLRE are significantly better than those of several advanced FE algorithms, and the improvement rate ranges from 2% to 15%.

Recurrent neural network-based speech recognition using MATLAB

The purpose of this paper is to design an efficient recurrent neural network (RNN)-based speech recognition system using software with long short-term memory (LSTM). The design process involves speech acquisition, pre-processing, feature extraction, training and pattern recognition tasks for a spoken sentence recognition system using LSTM-RNN. There are five layers namely, an input layer, a fully connected layer, a hidden LSTM layer, SoftMax layer and a sequential output layer. A vocabulary of 80 words which constitute 20 sentences is used. The depth of the layer is chosen as 20, 42 and 60 and the accuracy of each system is determined. The results reveal that the maximum accuracy of 89% is achieved when the depth of the hidden layer is 42. Since the depth of the hidden layer is fixed for a task, increased performance can be achieved by increasing the number of hidden layers.

Classification of Mangosteen Surface Quality Using Principal Component Analysis

Mangosteen (Garcinia mangostana L) is one of the primary contributor for Indonesia export. For export commodity, the fruit should comply the quality requirement including its surface. Presently, the surface is evaluated by human visual to classify between defect and non- defect surface. This conventional method is less accurate and takes time, especially in high volume harvest. In order to overcome this problem, this research proposed images processing based classification method using principal component analysis (PCA). The method involved pre-processing task, PCA decomposition, and statistical features extraction and classification task using linear discriminant analysis. The method has been tested on 120 images by applying 4-fold cross validation method and achieve classification accuracy of 96.67%, 90.00%, 90.00% and 100.00% for fold-1, fold-2, fold-3 and fold-4, respectively. In conclusion, the proposed method succeeded to classify between defect and non-defect mangosteen surface with 94.16% accuracy.

Recurrent neural network-based speech recognition using MATLAB

The purpose of this paper is to design an efficient recurrent neural network (RNN)-based speech recognition system using software with long short-term memory (LSTM). The design process involves speech acquisition, pre-processing, feature extraction, training and pattern recognition tasks for a spoken sentence recognition system using LSTM-RNN. There are five layers namely, an input layer, a fully connected layer, a hidden LSTM layer, SoftMax layer and a sequential output layer. A vocabulary of 80 words which constitute 20 sentences is used. The depth of the layer is chosen as 20, 42 and 60 and the accuracy of each system is determined. The results reveal that the maximum accuracy of 89% is achieved when the depth of the hidden layer is 42. Since the depth of the hidden layer is fixed for a task, increased performance can be achieved by increasing the number of hidden layers.

Nonpeaked Discriminant Analysis for Data Representation.

Of late, there are many studies on the robust discriminant analysis, which adopt L1-norm as the distance metric, but their results are not robust enough to gain universal acceptance. To overcome this problem, the authors of this article present a nonpeaked discriminant analysis (NPDA) technique, in which cutting L1-norm is adopted as the distance metric. As this kind of norm can better eliminate heavy outliers in learning models, the proposed algorithm is expected to be stronger in performing feature extraction tasks for data representation than the existing robust discriminant analysis techniques, which are based on the L1-norm distance metric. The authors also present a comprehensive analysis to show that cutting L1-norm distance can be computed equally well, using the difference between two special convex functions. Against this background, an efficient iterative algorithm is designed for the optimization of the proposed objective. Theoretical proofs on the convergence of the algorithm are also presented. Theoretical insights and effectiveness of the proposed method are validated by experimental tests on several real data sets.

Hierarchical Picture of existing Audio-Visual Speech Database

Despite the technological improvement and arrival of new methodologies in the different process of a speech-based applications, a parallel development is not observed in the availability of audio-visual speech database. This paper provides a detailed hierarchical picture of the existing audio-visual speech database. Since the performance of a speech-based application deeply depends on the different parameters like the number of speakers, speaker variability, phonetically balanced sentences, recording quality etc. involved in the creation of a database to attain specific task. This paper gave more importance to these parameters involved in the exciting audio-visual speech database rather than the experimental side which need linguistic knowledge about the concerned language in the feature extraction task and classification task. This paper is arranged in such a way that a new face in this realm can capture the needy things to build a speech database in his language. In addition, this paper differs from other review papers in the aspect that it gives equal importance to the available audio-visual speech database in the resourced and under-resourced languages.

Learning Latent Low-Rank and Sparse Embedding for Robust Image Feature Extraction

To defy the curse of dimensionality, the inputs are always projected from the original high-dimensional space into the target low-dimension space for feature extraction. However, due to the existence of noise and outliers, the feature extraction task for corrupted data is still a challenging problem. Recently, a robust method called low rank embedding (LRE) was proposed. Despite the success of LRE in experimental studies, it also has many disadvantages: 1) The learned projection cannot quantitatively interpret the importance of features. 2) LRE does not perform data reconstruction so that the features may not be capable of holding the main energy of the original "clean" data. 3) LRE explicitly transforms error into the target space. 4) LRE is an unsupervised method, which is only suitable for unsupervised scenarios. To address these problems, in this paper, we propose a novel method to exploit the latent discriminative features. In particular, we first utilize an orthogonal matrix to hold the main energy of the original data. Next, we introduce an l2,1 -norm term to encourage the features to be more compact, discriminative and interpretable. Then, we enforce a columnwise l2,1 -norm constraint on an error component to resist noise. Finally, we integrate a classification loss term into the objective function to fit supervised scenarios. Our method performs better than several state-of-the-art methods in terms of effectiveness and robustness, as demonstrated on six publicly available datasets.

Convolution Neural Network: A Shallow Dive in to Deep Neural Net Technology

It is always beneficial to reassess the previously done work to create interest and develop understanding about the subject in importance. In computer vision, to perform the task of feature extraction, classification or segmentation, measurement and assessment of image structures (medical images, natural images etc.) is to be done very efficiently. In the field of image processing numerous techniques are available, but it is very difficult to perform these tasks due to noise and other variable artifacts. Various Deep machine learning algorithms are used to perform complex task of recognition and computer vision. Recently Convolutional Neural Networks (CNNs-back bone of numerous deep learning algorithms) have shown state of the art performance in high level computer vision tasks, such as object detection, object recognition, classification, machine translation, semantic segmentation, speech recognition, scene labelling, medical imaging, robotics and control, , natural language processing (NLP), bio-informatics, cybersecurity, and many others. Convolution neural networks is the attempt to combine mathematics to computer science with icing of biology on it. CNNs work in two parts. The first part is mathematics that supports feature extraction and second part is about classification and prediction at pixel level. This review is intended for those who want to grab the complete knowledge about CNN, their development form ancient age to modern state of art system of deep learning system. This review paper is organized in three steps: in the first step introduction about the concept is given along with necessary background information. In the second step other highlights and related work proposed by various authors is explained. Third step is the complete layer wise architecture of convolution networks. The last section is followed by detailed discussion on improvements, and challenges on these deep learning techniques. Most papers consider for this review are later than 2012 from when the history of convolution neural networks and deep learning begins

Ground target localization and recognition via descriptors fusion

Keypoint matching can be defined as locating the position of a particular point in two images precisely. Recently, keypoint descriptors have taken a great effect targeting to be powerfully invariant to rotation, scale and translation for improving target detection. The detection task is carried out by a reference-scene image matching to localize the desired target in the input scene. An innovative approach is proposed in this work to fuse the state-of-the-art feature descriptors ORB, BRISK for the sake of accurate ground target detection in two phases. Firstly, off-line phase, where the fused features are extracted from different perspective, azimuth angles of the desired target to build a comprehensive reference image representation. Secondly, on-line phase, where the fused features extraction task is carried out from the whole scene. Hence, it is matched with the stored reference one to find the keypoints correspondence. The outliers’ problem is eliminated using Random Sample Consensus (RANSAC) algorithm resulting in speeding up the matching procedure. The conducted comparative analysis has revealed the discriminative power of the fused features in localization and recognition tasks while keeping the proposed system works in real-time.

Rough-Wavelet Feature Space, Deep Autoencoder, and Hyperspectral Image Classification

Prime objective of this letter is to select the most relevant features from the original set and perform two steps of feature extraction operations on those selected set, for the classification of hyperspectral remote sensing (HSRS) images. Neighborhood rough sets (NRSs)-based method is used for feature selection because of its excellent neighboring information capturing ability. On these selected features, two steps of extraction operations are performed using the stationary wavelet transform (WT) and stacked deep autoencoder (SDAE). Stationary WT extracts the features by exploiting the spectral–spatial information and stacked DAE extracts through representative learning of input information. The wavelet features and the original input spectral features are cascaded to feed as input to the stacks DAE for feature extraction and classification tasks. The proposed classification model with these operational steps possesses the ability to capture more informative features with improved spectral–spatial information that are highly beneficial for the classification of complex data sets, like HSRS images. Simulation results with two HSRS images justified the efficacy of the proposed model compared to other similar methods in terms of different performance measurement indexes.

A novel feature-based probability of detection assessment and fusion approach for reliability evaluation of vibration-based diagnosis systems

Structural health monitoring systems are based on suitable sensor techniques allowing online and offline supervision of technical systems. The quantification of sensors/measurement devices is a key issue for qualifying their effectiveness and efficiency and therefore to ensure safe operations. Probability of detection serves as a performance measure for quantifying the reliability of conventional nondestructive testing procedures taking into account statistical variability of sensor-based measurements. For vibration-based supervision approaches and fault detection and isolation methods, the probability of detection approach cannot be applied similarly. This results mainly from the complexity of the dynamical behavior of systems monitored in relation to faults, sensors position (observability), and the related feature extraction or monitoring task. In this contribution, probability of detection evaluation of vibration-based fault detection of elastic mechanical structures to be monitored is developed. Beside a principal discussion of the problem serving as introduction, an example using different sensor types in combination with mechanical modifications of an elastic beam is presented. The a90/95-criteria representing probability of 90% detection at a confidence level of 95% is examined to the measurements and related outcomes. Based on the analysis of a suitably chosen feature (like eigenfrequency or band power) and dependent on the mechanical modes considered, the efficiency and deficiency of the different combinations are shown. Based on the proposed approach, a new insight into the usefulness of different sensor type and fault position combination becomes possible. To improve the detection quality, suitable assumptions in combination with sensor/information fusion are applied to feature-based analysis as detection task using vibration measurements. In addition, based on an experimental evaluation, it can be concluded that a suitable fault-feature probability of detection analysis can be successfully implemented as a new reliable measure for vibration-based fault detection and isolation approaches. Furthermore, decision fusion as the combination of different measurements will allow the improvement of results. Dependent on noise analysis, a trade-off between flaw size detection and probability of falsely characterizing a fault with a90/95 reliability level can be attained.

Comparing ensemble strategies for deep learning: An application to facial expression recognition

Recent works have shown that Convolutional Neural Networks (CNNs), because of their effectiveness in feature extraction and classification tasks, are suitable tools to address the Facial Expression Recognition (FER) problem. Further, it has been pointed out how ensembles of CNNs allow improving classification accuracy. Nevertheless, a detailed experimental analysis on how ensembles of CNNs could be effectively generated in the FER context has not been performed yet, although it would have considerable value for improving the results obtained in the FER task. This paper aims to present an extensive investigation on different aspects of the ensemble generation, focusing on the factors that influence the classification accuracy on the FER context. In particular, we evaluate several strategies for the ensemble generation, different aggregation schemes, and the dependence upon the number of base classifiers in the ensemble. The final objective is to provide some indications for building up effective ensembles of CNNs. Specifically, we observed that exploiting different sources of variability is crucial for the improvement of the overall accuracy. To this aim, pre-processing and pre-training procedures are able to provide a satisfactory variability across the base classifiers, while the use of different seeds does not appear as an effective solution. Bagging ensures a high ensemble gain, but the overall accuracy is limited by poor-performing base classifiers. The impact of increasing the ensemble size specifically depends on the adopted strategy, but also in the best case the performance gain obtained by involving additional base classifiers becomes not significant beyond a certain limit size, thus suggesting to avoid very large ensembles. Finally, the classic averaging voting proves to be an appropriate aggregation scheme, achieving accuracy values comparable to or slightly better than the other experimented operators.

A Novel Convolution Computing Paradigm Based on NOR Flash Array With High Computing Speed and Energy Efficiency

Convolution is one of the key operations in signal processing and machine learning applications. In this paper, we propose a novel convolution computing paradigm based on the NOR Flash array (NFA) that is capable of executing the 2-D convolution computing in one clock cycle. In order to demonstrate the feasibility and efficiency of the proposed convolution computing paradigm, the feature extraction task on an image with the size of $20\times 20$ is executed using the NFA structure. We also prove the NOR Flash-driven convolution computing is capable of processing the image with a larger size. This paper presents a new approach to realize convolution computing with high speed and energy efficiency for the signal processing and convolution neural network.

An application of convolutional neural networks with salient features for relation classification

BackgroundDue to the advent of deep learning, the increasing number of studies in the biomedical domain has attracted much interest in feature extraction and classification tasks. In this research, we seek the best combination of feature set and hyperparameter setting of deep learning algorithms for relation classification. To this end, we incorporate an entity and relation extraction tool, PKDE4J to extract biomedical features (i.e., biomedical entities, relations) for the relation classification. We compared the chosen Convolutional Neural Networks (CNN) based classification model with the most widely used learning algorithms.ResultsOur CNN based classification model outperforms the most widely used supervised algorithms. We achieved a significant performance on binary classification with a weighted macro-average F1-score: 94.79% using pre-extracted relevant feature combinations. For multi-class classification, the weighted macro-average F1-score is estimated around 86.95%.ConclusionsOur results suggest that our proposed CNN based model using the not only single feature as the raw text of the sentences of biomedical literature, but also coupling with multiple and highlighted features extracted from the biomedical sentences could improve the classification performance significantly. We offer hyperparameter tuning and optimization approaches for our proposed model to obtain optimal hyperparameters of the models with the best performance.

Using Multiple-Feature-Spaces-Based Deep Learning for Tool Condition Monitoring in Ultraprecision Manufacturing

Tool condition monitoring is critical in ultraprecision manufacturing in order to optimize the performance of the overall process, while maintaining the desired part quality. Recently, deep learning has been successfully applied to numerous classification tasks in manufacturing, often to forecast part quality. In this paper, a novel deep learning data-driven modeling framework is presented, which includes a fusion of multiple stacked sparse autoencoders for tool condition monitoring in ultraprecision machining. The proposed computational framework consists of two main structures. First, a training model that is designed with the ability to process multiple parallel feature spaces to learn the lower-level features. Second, a feature fusion structure that is used to learn the higher-level features and associations to tool wear. To achieve this learning structure, a modified loss function is utilized that enhances the feature extraction and classification tasks. A dataset from a real manufacturing process is used to demonstrate the performance of the proposed framework. Experimental results and simulations show that the proposed method successfully classifies the ultraprecision machining case study with over 96% accuracy, while also outperforming comparable methodologies.

Research on FPGA filtering method of Gearbox fault signal

Gearboxes are an important part of marine machinery and equipment such as steam turbines and generators. Under normal circumstances, the performance of the gearbox directly affects whether the relevant mechanical equipment can normally work, so it is very important to detect and eliminate the gearbox failure. The gearbox vibration signal contains a wealth of working status information, and the use of gearbox vibration signal diagnosis is a common method. Since the vibration signal contains strong noise, denoising is the primary task of fault feature extraction. Aiming at this practical problem, in this paper, a denoising filter based on FPGA environment is designed. This filter is used to process the simulated gear fault signal with a signal-to-noise ratio (SNR) of 0.1206. By comparing the time domain diagram and frequency domain diagram before and after processing, the feasibility of using FPGA to realize the mechanical vibration signal processing is verified. This is of great practical significance to the miniaturization and real-time of the mechanical fault diagnosis system.

A novel electrocardiogram feature extraction approach for cardiac arrhythmia classification

In this work, we propose a novel approach to detect cardiac arrhythmias in electrocardiograms (ECG). The proposal focuses on different feature extractors and machine learning methods. The feature extraction techniques evaluated were Fourier, Goertzel, Higher Order Statistics (HOS), and Structural Co-Occurrence Matrix (SCM). As far as the authors know, this is the first time that SCM has been applied to the feature extraction task with ECG signals. Four well-known classifiers, commonly referred to in the literature (Support Vector Machine, Multi-Layer Perceptron, Bayesian, and Optimum-Path Forest) were tested and we compared our results with six classical feature extraction methods. Furthermore, the Association for the Advancement of Medical Instrumentation protocol was adopted and we made use of the MIT-BIH Arrhythmia Database for producing reliable results for clinical analysis. The confidence level to identify heart dysrhythmia in our results was 2% greater than other approaches in the literature. The proposed system is 1.3% more accurate than the best approach reported to date, and is 106 times faster. blackIn short, it is clinical reliable to use HOS for describing types of arrhythmia, since achieved 94.3% of accuracy.

Traffic-flow analysis for source-side DDoS recognition on 5G environments

This paper introduces a novel approach for detecting the participation of a protected network device in flooding-based Distributed Denial of Service attacks. With this purpose, the traffic flows are inspected at source-side looking for discordant behaviors. In contrast to most previous solutions, the proposal assumes the non-stationarity and heterogeneity inherent in the emergent communication environment. In particular, the approach takes advantage of the monitorization and knowledge acquisition capabilities implemented in the SELFNET (H2020-ICT-2014-2/671672) project, which facilitates its implementation as a self-organizing solution on 5G mobile networks. Monitorization, feature extraction and knowledge acquisition tasks are carried out on centralized control plane, hence the proposed architecture minimizes the impact on operational performance and prompts the end-points mobility. The preliminary results observed when considering different metrics, adjustment parameters, and a dataset with traffic observed in 61 real devices proven efficiency when distinguishing normal activities from DDoS behaviors of different intensity. With an optimal granularity selection, the highest AUC reached values close to 1.0 when measured under the most intense attacks, hence demonstrating optimal TPR and FPR relationships by adapting to the instantiated use cases.

Automatic crack detection for tunnel inspection using deep learning and heuristic image post-processing

In this paper, a crack detection mechanism for concrete tunnel surfaces is presented. The proposed methodology leverages deep Convolutional Neural Networks and domain-specific heuristic post-processing techniques to address a variety of challenges, including high accuracy requirements, low operational times and limited hardware resources, poor and variable lighting conditions, low textured lining surfaces, scarcity of training data, and abundance of noise. The proposed framework leverages the representational power of the convolutional layers of CNNs, which inherently selects effective features, thus obviating the need for the tedious task of handcrafted feature extraction. Additionally, the good performance rates attained by the proposed framework are acquired at a significantly lower execution time compared to other techniques. The presented mechanism was designed and developed as a core component of an autonomous robotic inspector deployed and validated in the tunnels of Egnatia Motorway in Metsovo, Greece. The obtained results denote the proposed approach’s superiority over a variety of methods and suggest a promising potential as a driver of autonomous concrete-lining tunnel-inspection robots.