Integrated Feature Extraction Research Articles

A Method of Text Detection and Recognition from Receipt Images Based on CRAFT and CRNN

Text detection and recognition from paper bank receipts image has a lot of applications in the business field, such as in power system marketing, which is applied in verification and cancellation of electricity charges and automatic archiving. In this paper, we use a text detection method to effectively detect text area in receipt image by exploring each character and affinity between characters. Then put the text sequence into text recognizition model which is combined by DCNN and RNN and integrates feature extraction, sequence modeling and transcription. The experiments demonstrate the superiority of the proposed algorithm over the prior arts, which performs well in the task of blurred image-based bank receipt chinese text recognition.

Multi-atlas segmentation of optic disc in retinal images via convolutional neural network

Multi-atlas segmentation is widely accepted as an essential image segmentation approach. Through leveraging on the information from the atlases instead of utilizing the model-based segmentation techniques, the multi-atlas segmentation could significantly enhance the accuracy of segmentation. However, label fusion, which plays an important role for multi-atlas segmentation still remains the primary challenge. Bearing this in mind, a deep learning-based approach is presented through integrating feature extraction and label fusion. The proposed deep learning architecture consists of two independent channels composing of continuous convolutional layers. To evaluate the performance our approach, we conducted comparison experiments between state-of-the-art techniques and the proposed approach on publicly available datasets. Experimental results demonstrate that the accuracy of the proposed approach outperforms state-of-the-art techniques both in efficiency and effectiveness.

Deep Ensemble Object Tracking Based on Temporal and Spatial Networks

In recent years, correlation filtering and deep learning have achieved good performance in object tracking. Correlation filtering is an efficient and real-time method because its formula provides a fast solution in the Fourier domain, but it does not benefit from end-to-end training. Although deep learning is an effective method for learning object representations, training deep networks online with one or a few examples is challenging. To address these problems, we propose a deep ensemble object tracking algorithm that fuses temporal and spatial information to improve algorithm precision and robustness. The framework of our algorithm includes four aspects: feature extraction, a baseline network, a branch network and adaptive ensemble learning. Feature extraction extracts the general object representation. The baseline network integrates feature extraction and a correlation filtering algorithm into a convolutional neural network for end-to-end training. The branch network is composed of a temporal network and a spatial network. The temporal and spatial networks capture the object temporal and spatial information and further refine the object position. Our algorithm only needs an initial frame to train all networks. Adaptive ensemble learning compensates for the object information deficiency and improves tracking accuracy. Many experiments on tracking benchmark datasets demonstrate that our algorithm performs favourably compared with state-of-the-art tracking algorithms.

Improvement in Hadoop performance using integrated feature extraction and machine learning algorithms

Big Data has been a term used in datasets which are complex and large in such a way there are some traditional technologies of data processing which are not adequate. Big Data can revolutionize most aspects in society such as collection or management of data from Big Data which is challenging and also very complex. The Hadoop has been designed for processing a large amount of unstructured and complex data. It has provided with a large amount of storage for data along with the ability to be able to tackle unlimited and concurrent tasks or jobs. The selection of features is an extremely powerful technique in the reduction of dimensionality and is also the most important step in machine learning applications. In recent decades, data is getting larger in a progressive manner in terms of instances and numbers making it very hard to deal with the problem of feature selection. In order to cope with such an epoch of Big Data, there are some more new techniques that are required to address the problem in a more efficient manner. At the same time, the suitability of the algorithms currently used may not be applicable especially when the size of data is above hundreds of gigabytes. For the purpose of this work, the correlation-based feature selection along with mutual information-based methods of feature selection was used for improving the performance. The AdaBoost and the support vector machine based classifiers have been used for improving their accuracy. The results of the experiment prove that the method proposed was able to achieve better performance compared to that of the other methods.

Enhancing Metabolome Coverage in Data-Dependent LC-MS/MS Analysis through an Integrated Feature Extraction Strategy.

In untargeted metabolomics, conventional data preprocessing software (e.g., XCMS, MZmine 2, MS-DIAL) are used extensively due to their high efficiency in metabolic feature extraction. However, these programs present limitations in recognizing low-abundance metabolic features, thus hindering complete metabolome coverage from the analysis. In this work, we explored the possibility of enhancing the metabolome coverage of data-dependent liquid chromatography-tandem mass spectrometry (LC-MS/MS) results by rescuing metabolic features that are missed by conventional software. To achieve this goal, we first categorized the metabolic features into four confidence levels based on their chromatographic peak shapes and the presence of corresponding MS/MS spectra. We then assessed the false positives and quantitative accuracy of the metabolic features that contain MS/MS spectra but are not recognized by conventional software. Our results indicate that these missed features contain valid and important metabolic information and should be integrated into the conventional metabolomics results. Thus, we developed a data-preprocessing pipeline to extract low-abundance metabolic features and integrate them with the results from conventional programs. This integrated feature extraction strategy was tested on a set of fecal metabolomic data retrieved from mice who have undergone normal diet vs high-fat diet treatments. In our test data set, the integrated feature extraction approach increased the number of significant features being extracted by 24.4% and identified five additional metabolites bearing critical biological meanings. Our results show that this integrated feature extraction strategy remarkably improves the metabolome coverage beyond that of conventional data preprocessing, therefore facilitating the confirmation of metabolites of interest and accomplishment of a higher success rate in de novo metabolite identification.

Capsule Network Assisted IoT Traffic Classification Mechanism for Smart Cities

With rapid development of compelling application scenarios of the Internet of Things (IoT), such as smart cities, it becomes substantially important to strengthen the management of data traffic in IoT networks. Traffic classification is beneficial in terms of both ensuring network security and improving quality of service. Traditional IoT traffic classification methods separate the classification algorithm and the design of feature engineering, which includes feature extraction and feature selection. Then, traffic identification or classification is performed by combining both. This paper proposes an end-to-end IoT traffic classification method relying on a deep learning aided capsule network for the sake of forming an efficient classification mechanism that integrates feature extraction, feature selection, and classification model. Our proposed traffic classification method beneficially eliminates the process of manually selecting traffic features, and is particularly applicable to smart city scenarios. To the best of our knowledge, this is the first time that capsule networks have been used in the context of traffic classification. Experimental results show the feasibility and effectiveness of our proposed traffic classification mechanism, which yields high classification accuracy.

Metandem: An online software tool for mass spectrometry-based isobaric labeling metabolomics

Mass spectrometry-based stable isotope labeling provides the advantages of multiplexing capability and accurate quantification but requires tailored bioinformatics tools for data analysis. Despite the rapid advancements in analytical methodology, it is often challenging to analyze stable isotope labeling-based metabolomics data, particularly for isobaric labeling using MS/MS reporter ions for quantification. We report Metandem, a novel online software tool for isobaric labeling-based metabolomics, freely available at http://metandem.com/web/. Metandem provides a comprehensive data analysis pipeline integrating feature extraction, metabolite quantification, metabolite identification, batch processing of multiple data files, online parameter optimization for custom datasets, data normalization, and statistical analysis. Systematic evaluation of the Metandem tool was demonstrated on UPLC-MS/MS, nanoLC-MS/MS, CE-MS/MS and MALDI-MS platforms, via duplex, 4-plex, 10-plex, and 12-plex isobaric labeling experiments and the application to various biological samples.

Applications of Stacked Autoencoder Network on Communication Transmitter Individual Feature Extraction

Aiming at the problem that square integral bispectrum cannot accurately represent communication transmitter individuals with the same type under small sample, a communication transmitter individual deep square integral bispectra feature extraction method based on stacked autoencoder network is proposed. The square integral bispectrum (SIB) features of real communication transmitters signals were extracted firstly from the instantaneous frequency. Then the deep SIB features were represented through the layer-wise semi-supervised learning by a stacked autoencoder network. Experiments on real FM communication transmitter signals showed that the identification accuracy is around 80% and the results of multiple experimental variables setting indicated the robustness of the algorithm proposed.

Dual-Input Neural Network Integrating Feature Extraction and Deep Learning for Coronary Artery Disease Detection Using Electrocardiogram and Phonocardiogram

Electrocardiogram (ECG) and phonocardiogram (PCG) signals reflect the electrical and mechanical activities of the heart, respectively. Although studies have documented that some abnormalities in ECG and PCG signals are associated with coronary artery disease (CAD), only few researches have combined the two signals for automatic CAD detection. This paper aims to differentiate between CAD and non-CAD groups using simultaneously collected ECG and PCG signals. To entirely exploit the underlying information in these signals, a novel dual-input neural network that integrates the feature extraction and deep learning methods is developed. First, the ECG and PCG features are extracted from multiple domains, and the information gain ratio is used to select important features. On the other hand, the ECG signal and the decomposed PCG signal (at four scales) are concatenated as a five-channel signal. Then, the selected features and the five-channel signal are fed into the proposed network composed of a fully connected model and a deep learning model. The results show that the classification performance of either feature extraction or deep learning is insufficient when using only ECG or PCG signal, and combining the two signals improves the performance. Further, when using the proposed network, the best result is obtained with accuracy, sensitivity, specificity, and G-mean of 95.62%, 98.48%, 89.17%, and 93.69%, respectively. Comparisons with existing studies demonstrate that the proposed network can effectively capture the combined information of ECG and PCG signals for the recognition of CAD.

PGC-Net: A Light Weight Convolutional Sequence Network for Digital Pressure Gauge Calibration

Automatic digital pressure gauge calibration is challenging due to various unconstrained conditions. Although existing CNN-RNN based methods have been almost perfect on scene text recognition, they fail to perform well on digital pressure gauge calibration that requires to be extremely computation-efficient and accurate. In this paper, we propose a light weight fully convolutional sequence recognition network for fast and accurate digital P ressure G auge C alibration (PGC-Net). PGC-Net integrates feature extraction, sequence modelling and transcription into a unified framework. Experimental results show that PGC-Net runs 28 fps on CPU with 97.41% accuracy. Compared with previous methods, PGC-Net achieves better or comparable performance at lower inference time. Without bells and whistles, PGC-Net is capable of recognizing decimal points that usually appear in pressure gauge images, which evidently verifies the feasibility of PGC-Net. We collected a dataset that contains 17, 240 gauge images with annotated labels for automatic digital pressure gauge calibration. The dataset has been public for future research.

Spatiotemporal Modeling for Video Summarization Using Convolutional Recurrent Neural Network

In this paper, a novel neural network named CRSum for the video summarization task is proposed. The proposed network integrates feature extraction, temporal modeling, and summary generation into an end-to-end architecture. Compared with previous work on this task, the proposed method owns three distinctive characteristics: 1) it for the first time leverages convolutional recurrent neural network for simultaneously modeling spatial and temporal structure of video for summarization; 2) thorough and delicate features of video are obtained in the proposed architecture by trainable three-dimension convolutional neural networks and feature fusion; and 3) a new loss function named Sobolev loss is defined, aiming to constrain the derivative of sequential data and exploit potential temporal structure of video. A series of experiments are conducted to prove the effectiveness of the proposed method. We further analyze our method from different aspects by well-designed experiments.

Student dropout prediction in massive open online courses by convolutional neural networks

Massive open online courses (MOOCs) have given global learners access to quality educational resources, but the persistent high dropout rates problem has a serious impact on their educational effectiveness. Therefore, how to predict the dropout in MOOCs and make advance intervention is a hot topic in the research of MOOCs in recent years. Traditional methods rely on handcrafted features, the workload is heavy, and it is difficult to ensure the final prediction effect. In order to solve this problem, this paper proposes an end-to-end dropout prediction model based on convolutional neural networks to predict the student dropout problem in MOOCs and it integrates feature extraction and classification into a single framework, which transforms the original timestamp data according to different time windows and automatically extracts features to achieve better feature representation. Extensive experiments on a public dataset show that our approach can achieve results comparable to other dropout prediction methods on precision, recall, F1 score, and AUC score.

Research of the Vibration Source Tracking in Phase-Sensitive Optical Time-Domain Reflectometry Signals Based by Image Processing Method

This paper aims to improve the source tracking efficiency of distributed vibration signals generated by phase-sensitive optical time-domain reflectometry (Φ-OTDR). Considering the two dimensions (time and length) of Φ-OTDR signals, the authors saved and processed these signals as images after particle filtering. The filtering method could save 0.1% of hard drive space without sacrificing the original features of the signals. Then, an integrated feature extraction method was proposed to further process the generated image. The method combines three individual extraction methods, namely, texture feature extraction, shape feature extraction and intrinsic feature extraction. Subsequently, the signal of each frame image was recognized to track the vibration source. To verify the effect of the proposed method, several experiments were carried out to compare it with popular and traditional approaches. The results show that: Hard drive space is greatly conserved by saving the distributed vibration signals as images; the proposed particle filter is a desirable way to screen the vibration signals for monitoring; the integrated feature extraction outperforms the individual extraction methods for texture features, shape features and intrinsic features; the proposed method has a better effect than other popular integrated feature extraction methods; and, the signal source tracking method has little impact on the positioning accuracy of the vibration source. The research findings provide important insights into the source tracking of Φ-OTDR signals.

Klasifikasi Teh Hijau dan Teh Hitam Tambi-Pagilaran dengan Metode Principal Component Analysis (PCA) Menggunakan E-Nose

This research aims to classification of samples of green tea and black tea originated from different planting sites, Tambi and Pagilaran. Samples of green tea and black tea; quality I (BOP), quality II (BP), quality III (Bohea) were each collected from Tambi and Pagilaran to analyze the charasteristic of both sample from both sites. Measurements of tea samples were performed using a dynamic e-nose device based on a MOS gas sensor, with a maximum set point temperature of 40ºC, flushing 300 seconds, collecting 120 seconds, and purging 80 seconds for 10 cycles repeatedly. The resulting sensor response is then processed using the difference method for baseline manipulation. Characteristic of extraction process on the sensor response results is carried out in three methods; relative, fractional change, and integral. Matrix data of the feature extraction results was reduced using the PCA method by mapping the aroma patterns of each sample using 2-PCA components. The PCA reduction results in integral feature extraction showed the largest percentage of cumulative variance in classifying green tea sample data by 97% and black tea by 100%. The large percentage value of cumulative variance indicates PCA can differentiate samples of green tea and black tea from Tambi and Pagilaran well.

Deep Coupling Autoencoder for Fault Diagnosis With Multimodal Sensory Data

Effective fault diagnosis of rotating machinery has multifarious benefits, such as improved safety, enhanced reliability, and reduced maintenance cost, for complex engineered systems. With many kinds of installed sensors for conducting fault diagnosis, one of the key tasks is to develop data fusion strategies that can effectively handle multimodal sensory signals. Most traditional methods use hand-crafted statistical features and then combine these multimodal features simply by concatenating them into a long vector to achieve data fusion. The present study proposes a deep coupling autoencoder (DCAE) model that handles the multimodal sensory signals not residing in a commensurate space, such as vibration and acoustic data, and integrates feature extraction of multimodal data seamlessly into data fusion for fault diagnosis. Specifically, a coupling autoencoder (CAE) is constructed to capture the joint information between different multimodal sensory data, and then a DCAE model is devised for learning the joint feature at a higher level. The CAE is developed by coupling hidden representations of two single-modal autoencoders, which can capture the joint information from multimodal data. The performance of the proposed method is evaluated by two experiments, which shows that the DCAE model succeeds in efficiently utilizing multisource sensory data to perform accurate fault diagnosis. Compared with other methods, the proposed method exhibits better performance.

Approximate Low-Rank Projection Learning for Feature Extraction.

Feature extraction plays a significant role in pattern recognition. Recently, many representation-based feature extraction methods have been proposed and achieved successes in many applications. As an excellent unsupervised feature extraction method, latent low-rank representation (LatLRR) has shown its power in extracting salient features. However, LatLRR has the following three disadvantages: 1) the dimension of features obtained using LatLRR cannot be reduced, which is not preferred in feature extraction; 2) two low-rank matrices are separately learned so that the overall optimality may not be guaranteed; and 3) LatLRR is an unsupervised method, which by far has not been extended to the supervised scenario. To this end, in this paper, we first propose to use two different matrices to approximate the low-rank projection in LatLRR so that the dimension of obtained features can be reduced, which is more flexible than original LatLRR. Then, we treat the two low-rank matrices in LatLRR as a whole in the process of learning. In this way, they can be boosted mutually so that the obtained projection can extract more discriminative features. Finally, we extend LatLRR to the supervised scenario by integrating feature extraction with the ridge regression. Thus, the process of feature extraction is closely related to the classification so that the extracted features are discriminative. Extensive experiments are conducted on different databases for unsupervised and supervised feature extraction, and very encouraging results are achieved in comparison with many state-of-the-arts methods.

Image-Based Process Monitoring Using Deep Belief Networks

With the advances in optical sensing and image capture systems, process images certainly offer new perspectives to process monitoring. Compared to the process data collected by traditional sensors at local regions, process images, which can capture more significant variations in the whole space, enhance the monitoring performance in data-driven monitoring methods. In this paper, a popular deep learning method, namely deep belief network (DBN), is applied to effectively extract useful features from the images. Meanwhile, a new statistic is developed for the DBN model, which integrates feature extraction and fault detection into one model rather than separately accomplish them. The effectiveness of the proposed DBN based monitoring method is demonstrated in a real combustion system.

A novel data mining approach for detection of brain disorder diseases using an integrated wavelet transform technique

An integrated form of wavelet transform technique (IDWT) has been explored and established as a feature extraction strategy to detect the brain disorder from brain MRI images is this study. There are different orientations of human brain images such as; sagittal, coronal and transaxial plane. Here, transaxial plane human brain images are collected and the advantages of Haar and Bi-orthogonal 1.3 wavelet functions are used to extract the relevant features in two-dimensional space. After features are extracted from two different integrated wavelet functions, then the entire combined feature matrix has been provided as an input to Support Vector Machine (SVM) with linear and polynomial kernel to classify different brain disorder diseases. This proposed integrated feature extraction strategy has been compared with the multi-layer probabilistic neural network (MLPNN), Naive Bayesian network and Logistic Regression as well as existing techniques studied during the literature survey and it has been observed that proposed DWT with SVM polynomial kernel is giving 100% result, additionally, the proposed model has been validated using accuracy, precision, recall and F1-score.

Feature extraction method based on VMD and MFDFA for fault diagnosis of reciprocating compressor valve

Aiming at the nonlinearity, nonstationarity and multi-component coupling characteristics of reciprocating compressor vibration signals, an integrated feature extraction method based on the variational mode decomposition (VMD) and multi-fractal detrended fluctuation analysis (MFDFA) is proposed for a fault diagnosis for a reciprocating compressor valve. Firstly, to eliminate the noise interference, a novel VMD method with superior anti-interference performance was utilized to obtain several components of the quasi-orthogonal band-limited intrinsic mode function (BLIMF) from a strong non-stationarity vibration signal, and a consistent number K of BLIMFs was selected based on a novel criterion for all fault states. Secondly, the MFDFA method, which can describe the multi-fractal structure feature of non-stationary time series, was applied to analyze each BLIMF component, and the parameters of MFDFA were employed as the eigenvectors to reflect the structure characteristics and local scale behavior of the vibration signal. Then, the principal component analysis (PCA) was introduced to refine the eigenvectors for a higher recognition efficiency and accuracy. Finally, the vibration signals of four types of reciprocating compressor valve faults were analyzed by this method, and the faults were identified correctly by pattern classifiers of BTSVM and CNN. Further results comparison with other feature extraction methods verifies the superiority of the proposed method.

Deteksi Daging Sapi Menggunakan Electronic Nose Berbasis Bidirectional Associative Memory

E-nose is an instrument used to detect odor. E-nose developed with Bidirectional Associative memory (BAM) algorithm has advantages in processing incomplete input data and noise. The purpose of the study was to implement the BAM algorithm to detect pure beef among samples of beef, pork, and mixed meat from aroma with e-nose.Data processing of the sample reading results begins by performing the baseline manipulation process, then do difference and integral feature extraction for the data. The characteristic extraction data will be converted into bipolar matrix patterns (1 and -1) so that the threshold data is needed to be able to determine the feature extraction data to be bipolar. Data that have become bipolar matrices will be used as test and reference data in the program with cross validation testing to obtain the percentage of truth of meat detection using BAM based e-nose.Detection of meat with BAM using integral feature extraction with bipolar the first way yields a 14,8% success percentage and the second way bipolar yields a 15,7% success rate. The extraction of characteristic difference with bipolar the first way yields a success percentage of 17,3% and the second way bipolar yields a success rate of 16,4%.