Generate Feature Vectors Research Articles

Transfer Learning for Feature Dimensionality Reduction

Transfer learning is a machine learning methodology by which a model developed for achieving a task is exploited for another related job. Many pre-trained image classification models trained on ImageNet are used for transfer learning. These pre-trained networks could also be used for classifying out of domain images by retraining them. This paper, along with the existing application for these pre-trained models, is also being exploited for feature dimensionality reduction. Many dimensionality reduction methods are available; the pre-trained image models will help us perform both image feature extraction and dimensionality reduction in a single go using the same network. The fine-tuning of the fully connected layers of the pre-trained network is done to extract the image features; along with this fine-tuning, some more tweaking is done on the fully connected layers of these models to reduce the image feature dimensionality. Here, VGG-16 and VGG-19 are the pre-trained models considered for feature vector generation and dimensionality reduction. An analysis of the efficiency of features generated by these pre-trained networks in classifying the out-of-domain images is done. Three different variants of VGG-16 and VGG-19 are analysed. All the three variants developed gave an AUC value above 0.8, which is considered good.

Leveraging Arabic sentiment classification using an enhanced CNN-LSTM approach and effective Arabic text preparation

The high variety in the forms of the Arabic words creates significant complexity related challenges in Natural Language Processing (NLP) tasks for Arabic text. These challenges can be dealt with by using different techniques for semantic representation, such as word embedding methods. In addition, approaches for reducing the diversity in Arabic morphologies can also be employed, for example using appropriate word normalisation for Arabic texts. Deep learning has proven to be very popular in solving different NLP tasks in recent years as well. This paper proposes an approach that combines Convolutional Neural Networks (CNNs) with Long Short-Term Memory (LSTM) networks to improve sentiment classification, by excluding the max-pooling layer from the CNN. This layer reduces the length of generated feature vectors after convolving the filters on the input data. As such, the LSTM networks will receive well-captured vectors from the feature maps. In addition, the paper investigated different effective approaches for preparing and representing the text features in order to increase the accuracy of Arabic sentiment classification.

A deep learning ensemble approach to prioritize antiviral drugs against novel coronavirus SARS-CoV-2 for COVID-19 drug repurposing

The alarming pandemic situation of Coronavirus infectious disease COVID-19, caused by the severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2), has become a critical threat to public health. The unexpected outbreak and unrealistic progression of COVID-19 have generated an utmost need to realize promising therapeutic strategies to fight the pandemic. Drug repurposing-an efficient drug discovery technique from approved drugs is an emerging tactic to face the immediate global challenge. It​ offers a time-efficient and cost-effective way to find potential therapeutic agents for the disease. Artificial Intelligence-empowered deep learning models enable the rapid identification of potentially repurposable drug candidates against diseases. This study presents a deep learning ensemble model to prioritize clinically validated anti-viral drugs for their potential efficacy against SARS-CoV-2. The method integrates the similarities of drug chemical structures and virus genome sequences to generate feature vectors. The best combination of features is retrieved by the convolutional neural network in a deep learning manner. The extracted deep features are classified by the extreme gradient boosting classifier to infer potential virus–drug associations. The method could achieve an AUC of 0.8897 with 0.8571 prediction accuracy and 0.8394 sensitivity under the fivefold cross-validation. The experimental results and case studies demonstrate the suggested deep learning ensemble system yields competitive results compared with the state-of-the-art approaches. The top-ranked drugs are released for further wet-lab researches.

PrimePatNet87: Prime pattern and tunable q-factor wavelet transform techniques for automated accurate EEG emotion recognition

Nowadays, many deep models have been presented to recognize emotions using electroencephalogram (EEG) signals. These deep models are computationally intensive, it takes a longer time to train the model. Also, it is difficult to achieve high classification performance using for emotion classification using machine learning techniques. To overcome these limitations, we present a hand-crafted conventional EEG emotion classification network. In this work, we have used novel prime pattern and tunable q-factor wavelet transform (TQWT) techniques to develop an automated model to classify human emotions. Our proposed cognitive model comprises feature extraction, feature selection, and classification steps. We have used TQWT on the EEG signals to obtain the sub-bands. The prime pattern and statistical feature generator are employed on the generated sub-bands and original signal to generate 798 features. 399 (half of them) out of 798 features are selected using minimum redundancy maximum relevance (mRMR) selector, and misclassification rates of each signal are evaluated using support vector machine (SVM) classifier. The proposed network generated 87 feature vectors hence, this model is named PrimePatNet87. In the last step of the feature generation, the best 20 feature vectors which are selected based on the calculated misclassification rates, are concatenated. The generated feature vector is subjected to the feature selection and the most significant 1000 features are selected using the mRMR selector. These selected features are then classified using an SVM classifier. In the last phase, iterative majority voting has been used to generate a general result. We have used three publicly available datasets, namely DEAP, DREAMER, and GAMEEMO, to develop our proposed model. Our presented PrimePatNet87 model reached over 99% classification accuracy on whole datasets with leave one subject out (LOSO) validation. Our results demonstrate that the developed prime pattern network is accurate and ready for real-world applications.

FILDNE: A Framework for Incremental Learning of Dynamic Networks Embeddings

Representation learning on graphs has emerged as a powerful mechanism to automate feature vector generation for downstream machine learning tasks. The advances in representation on graphs have centered on both homogeneous and heterogeneous graphs, where the latter presenting the challenges associated with multi-typed nodes and/or edges. In this paper, we consider the additional challenge of evolving graphs. We ask the question of whether the advances in representation learning for static graphs can be leveraged for dynamic graphs and how? It is important to be able to incorporate those advances to maximize the utility and generalization of methods. To that end, we propose the Framework for Incremental Learning of Dynamic Networks Embedding (FILDNE), which can utilize any existing static representation learning method for learning node embeddings while keeping the computational costs low. FILDNE integrates the feature vectors computed using the standard methods over different timesteps into a single representation by developing a convex combination function and alignment mechanism. Experimental results on several downstream tasks, over seven real-world datasets, show that FILDNE is able to reduce memory (up to 6x) and computational time (up to 50x) costs while providing competitive quality measure gains (e.g., improvements up to 19 pp AUC on link prediction and up to 33 pp mAP on graph reconstruction) with respect to the contemporary methods for representation learning on dynamic graphs.

Student Behavior Recognition System for the Classroom Environment Based on Skeleton Pose Estimation and Person Detection.

Human action recognition has attracted considerable research attention in the field of computer vision, especially for classroom environments. However, most relevant studies have focused on one specific behavior of students. Therefore, this paper proposes a student behavior recognition system based on skeleton pose estimation and person detection. First, consecutive frames captured with a classroom camera were used as the input images of the proposed system. Then, skeleton data were collected using the OpenPose framework. An error correction scheme was proposed based on the pose estimation and person detection techniques to decrease incorrect connections in the skeleton data. The preprocessed skeleton data were subsequently used to eliminate several joints that had a weak effect on behavior classification. Second, feature extraction was performed to generate feature vectors that represent human postures. The adopted features included normalized joint locations, joint distances, and bone angles. Finally, behavior classification was conducted to recognize student behaviors. A deep neural network was constructed to classify actions, and the proposed system was able to identify the number of students in a classroom. Moreover, a system prototype was implemented to verify the feasibility of the proposed system. The experimental results indicated that the proposed scheme outperformed the skeleton-based scheme in complex situations. The proposed system had a 15.15% higher average precision and 12.15% higher average recall than the skeleton-based scheme did.

Violence Detection over Online Social Networks: An Arabic Sentiment Analysis Approach

<p class="0abstract">Today, the influence of the social media on different aspects of our lives is increasing, many scholars from various disciplines and majors looking at the social media networks as the ongoing revolution. In Social media networks, many bonds and connections can be established whether being direct or indirect ties. In fact, Social networks are used not only by people but also by companies. People usually create their own profiles and join communities to discuss different common issues that they have interest in. On the other hand, companies also can create their virtual presence on the social media networks to benefit from this media to understand the customers and gather richer information about them. With all of the benefits and advantages of social media networks, they should not always be seen as a safe place for communicating, sharing information and ideas, and establishing virtual communities. These information and ideas could carry with them hatred speeches that must be detected to avoid raising violence. Therefore, web content mining can be used to handle this issue. Web content mining is gaining more concern because of its importance for many businesses and institutions. Sentiment Analysis (SA) is an important sub-area of web content mining. The purpose of SA is to determine the overall sentiment attitude of writer towards a specific entity and classify these opinions automatically. There are two main approaches to build systems of sentiment analysis: the machine learning approach and the lexicon-based approach. This research presents the design and implementation for violence detection over social media using machine learning approach. Our system works on Jordanian Arabic dialect instead of Modern Standard Arabic (MSA). The data was collected from two popular social media websites (Facebook, Twitter) and has used native speakers to annotate the data. Moreover, different preprocessing techniques have been used to show their effect on our model accuracy. The Arabic lexicon was used for generating feature vectors and separate them to features set. Here, we have three well known machine learning algorithms: Support Vector Machine (SVM), Naive Bayes (NB) and k-Nearest Neighbors (KNN). Building on this view, Information Science Research Institute’s (ISRI) stemming and stop word file as a result of preprocessing were used to extract the features. Indeed, several features have been extracted; however, using the SVM classifier reveals that unigram and features extracted from lexicon are characterized by the highest accuracy to detect violence.</p>

관주형 철탑 상태 감시를 위한 음향 방출 신호처리에 따른 특징 분석

In this study, we propose and analyze a machine learning method based on the genetic algorithm (GA) and supporting vector machine (SVM) for the effective classification of faults detected by an acoustic emission test on the welding parts of tubular steel towers. A band-pass filter, an envelope analysis (EA), and an intensified EA (IEA) are employed to generate feature vectors for the machine learning method based on the GA. After signal processing, the signals are applied to GA-based machine learning to derive the representative features of the received signal, and the SVM classifies the fault signals and normal signals from the detected signals. Consequently, it is confirmed that the received signal processed by EA and IEA can classify faults with an accuracy of 93 % or more. Hence, the proposed fault test and classification method is expected to be useful in the development of a system for constant monitoring and early detection of welding faults inside a tubular steel tower.

Fuzzy deep learning-based crop yield prediction model for sustainable agronomical frameworks

The evolution in science and innovation has to lead to an immense volume of information from various agricultural fields to be accumulated in the public domain. As a result, an objective arises from the investigation of the accessible information and incorporating them with processes like foreseeing crop yield, plant diseases examination, crops enhancement, etc. Machine learning has grown with tremendous processing methods to conceive new innovations in the multi-disciplinary agricultural sector. In experimenting with machine learning models, there exist certain limitations like improvident nonlinear mapping between the raw data and crop yield values. Hence, deep learning models are comprehensively used to extricate critical crop parameters for prediction. Foreseeing the crop yield depending on climate, soil and water parameters has been a potential research subject. This paper proposes a hybrid deep learning-based crop yield prediction system using deep belief network (DBN) and fuzzy neural networks system (FNN). DBN is a combination of statistics and probability with neural networks. Though DBN performs better for nonlinear systems, the algorithm alone cannot provide satisfactory results in terms of robustness, model accuracy and learning speed, which is predominantly due to gradient diffusion. Hence, a DBN along with FNN has been proposed to overcome the nonlinearity and gradient diffusion problems. The proposed model initially performs an efficient pre-training technique by DBN for enhanced model development and feature vector generation. This characteristic feature vector is fed as an input to the FNN for further processing. The superiority of the proposed fuzzy neural network-based deep belief network is analyzed by comparing it with other deep learning algorithms. The proposed model efficiently predicts the results outperforming the other models by preserving the original data distribution with an accuracy of 92%.

A joint change detection method on complex-valued polarimetric synthetic aperture radar images based on feature fusion and similarity learning

ABSTRACT Existing deep learning-based change detection methods in the field of polarimetric synthetic aperture radar (POLSAR) usually directly deal with intensity images. Methods can be easily transferred from optical image processing to synthetic aperture radar (SAR) image processing. However, the polarization information, which is critical in POLSAR data, has been discarded under normal situations. This paper introduces a novel joint change detection network based on similarity learning for coregistered complex-valued SAR data. A pseudo-siamese network takes both amplitude information and polarization information of POLSAR data as the input. The fusion of low-level, middle-level and high-level features enables the network to keep high-resolution and have strong representation ability during training procedures. Our novel sub-networks, which we term C2-Net and Intensity-Net, deal with the covariance matrix of complex SAR data and amplitude SAR data, respectively. The Intensity-Net works as a typical classification network and detects targets directly. The C2-Net attempts to find the relationship between two SAR data patches. An improved cosine similarity function is used to measure the similarity between two generated feature vectors in C2-Net. Output probability vectors of the two sub-networks are combined for final change detection. The two sub-networks are trained jointly and simultaneously. Control experiments show that proposed improvements are working. Experimental results on dual-polarization complex-valued SAR data from Sentinel-1 demonstrate the effectiveness of the proposed method.

An improved scheme for multifeature-based foreground detection using challenging conditions

Detecting an accurate foreground from a video frame is a critical task under different complex situations such as sudden changes in illuminations, relocation of background objects, shadow, low contrast videos and dynamic backgrounds (like waving tree, rippling of water etc.). Most of the existing schemes utilize a single feature-based foreground detection approach, which in turn is hard to apply under aforementioned complex situations. In order to mitigate this issue and properly exploit the different characteristics of the pixels, the present work proposes an efficient foreground detection scheme for better segmenting the foreground. In the proposed scheme, first the texture features are first extracted utilizing cross-diagonal texture matrix (CDTM), which essentially combines the merits of both the gray level co-occurrence matrix (GLCM) and the texture spectrum (TS) to provide a complete texture information about a frame. The color and gray value features of the pixel along with texture features are utilized for the feature vector generation. Second, during background modeling phase, the similarity distance measure is computed employing the Canberra distance between the mean feature vector of the current frame and the model. Finally, a method for adaptively selecting the threshold value is proposed, instead of choosing heuristically to correctly classify the foreground and background pixels under the dynamic background condition when background pixels changing frequently.Experiments are conducted using a wide variety of indoor and outdoor video sequences under various different challenging conditions and the results compared with that of the existing state-of-the-art methods. From the experimental results, it is shown that the proposed scheme outperforms the existing schemes in terms of the quantitative as well as qualitative measures.

Metric learning with generator for closed loop detection in VSLAM

The development of Driverless Car, Unmanned Aerial Vehicle, Human–Computer Interaction and Artificial Intelligence has promoted the Internet of Things (IoT) industry, in which, Visual Simultaneous Localization and Mapping (VSLAM) is an important Localization and Mapping technique. Closed loop detection can alleviate the error accumulation during the operation of VSLAM. The traditional closed loop detection methods mostly rely on manually defined features, subjective and unstable, which are difficult to cope with complex and repetitive scenarios. Thus, triplet loss-based metric learning has been considered as a better solution for closed loop detection. In this paper, first, constructed Generator is applied to generate feature vector of hard negative sample. Second, triplet loss and generative loss have been applied to construct loss function. The keyframes are converted into feature vectors with well-trained model, evaluating the similarity of keyframes by calculating their distance of feature vectors, which is used to determine whether a closed loop is formed. Finally, TUM dataset is introduced to evaluate the Precision and Recall of the proposed metric learning. The well-trained model is applied to establish loop closing thread for VSLAM system. The experimental results illustrate the feasibility and effectiveness of the metric learning-based closed loop detection, which can be further applied to practical VSLAM systems.

Design and implementation of convolutional neural network-based SVM technique for manufacturing defect detection

This paper introduces the design, implementation, training and testing of deep convolutional neural network (DCNN)-based support vector machines (SVMs). These DCNN-based SVMs are designed using software tools developed by the authors that enable them to construct, train, and test the DCNN-based SVMs and effectively facilitate vision-based inspection to detect different undesirable manufacturing defects. Two pretrained DCNNs are used for this purpose: the sssNet is developed by the authors and was trained using many actual and simple target images consisting of seven categories, and the standard AlexNet that was trained by a large number of images consisting of 1,000 categories. In this work, the pretrained sssNet and AlexNet are used as feature vector extractors in training and testing. The generated feature vectors are used as inputs to train SVMs for the final binary classification represented as accept (OK) or reject (NG) category.

Weighted Islanding Detection for DC Microgrid Based on Random Forest Classification

At present, the main form of microgrid is AC grid. DC microgrids have received extensive attention and research with the rapid development of various DC power. The operation mode of the DC microgrid is divided into grid-connected operation and islanding operation. Islanding is formed after the circuit breaker tripped, which connects microgrid to large grid. Islanding operation can be divided into planned islanding and unplanned islanding. Unplanned islanding will cause certain harm to users and systems, so it is necessary to detect islanding accurately in the DC microgrid. This paper proposes an islanding detection method for DC microgrid based on random forest classification. Firstly, raw data is cleaned, extracted features and generated feature vector set. The extracted features include six islanding characteristic indexes, which consist of voltage, current, output active power and their first order backward difference on the DC bus side. Then, based on random forest classification, building the islanding detection model. Islanding detection model for DC microgrid can distinguish islanding event successfully and accurately. Based on weighted random forest classification, it can detect islanding event more accurately compared with decision tree classification when processing large amounts of data.

Bionoi: A Voronoi Diagram-Based Representation of Ligand-Binding Sites in Proteins for Machine Learning Applications.

Bionoi is a new software to generate Voronoi representations of ligand-binding sites in proteins for machine learning applications. Unlike many other deep learning models in biomedicine, Bionoi utilizes off-the-shelf convolutional neural network architectures, reducing the development work without sacrificing the performance. When initially generating images of binding sites, users have the option to color the Voronoi cells based on either one of six structural, physicochemical, and evolutionary properties, or a blend of all six individual properties. Encouragingly, after inputting images generated by Bionoi into the convolutional autoencoder, the network was able to effectively learn the most salient features of binding pockets. The accuracy of the generated model is evaluated both visually and numerically through the reconstruction of binding site images from the latent feature space. The generated feature vectors capture well various properties of binding sites and thus can be applied in a multitude of machine learning projects. As a demonstration, we trained the ResNet-18 architecture from Microsoft on Bionoi images to show that it is capable to effectively classify nucleotide- and heme-binding pockets against a large dataset of control pockets binding a variety of small molecules. Bionoi is freely available to the research community at https://github.com/CSBG-LSU/BionoiNet.

Gabor and Maximum Response Filters with Random Forest Classifier for Face Recognition in the Wild

Research on face recognition has been evolving for decades. There are numerous approaches developed with highly desirable outcomes in constrained environments. In contrast, approaches to face recognition in an unconstrained environment where varied facial posing, occlusion, aging, and image quality still pose vast challenges. Thus, face recognition in the unconstrained environment still an unresolved problem. Many current techniques are not performed well when experimented in unconstrained databases. Additionally, most of the real-world application needs a good face recognition performance in the unconstrained environment. This paper presents a comprehensive process aimed to enhance the performance of face recognition in an unconstrained environment. This paper presents a face recognition system in an unconstrained environment. The fusion between Gabor filters and Maximum Response (MR) filters with Random Forest classifier is implemented in the proposed system. Gabor filters are a hybrid of Gabor magnitude filters and Oriented Gabor Phase Congruency (OGPC) filters. Gabor magnitude filters produce the magnitude response while the OGPC filters produce the phase response of Gabor filters. The MR filters contain the edge- and bar-anisotropic filter responses and isotropic filter responses. In the face features selection process, Monte Carlo Uninformative Variable Elimination Partial Least Squares Regression (MC-UVE-PLSR) is used to select the optimal face features in order to minimize the computational costs without compromising the accuracy of face recognition. Random Forests is used in the classification of the generated feature vectors. The algorithm performance is evaluated using two unconstrained facial image databases: Labelled Faces in the Wild (LFW) and Unconstrained Facial Images (UFI). The proposed technique used produces encouraging results in these evaluated databases in which it recorded face recognition rates that are comparable with other state-of-the-art algorithms.

Detection and tracking of cracks based on thermoelastic stress analysis.

Thermoelastic stress analysis using arrays of small, low-cost detectors has the potential to be used in structural health monitoring. However, evaluation of the collected data is challenging using traditional methods, due to the lower resolution of these sensors, and the complex loading conditions experienced. An alternative method has been developed, using image decomposition to generate feature vectors which characterize the uncalibrated map of the magnitude of the thermoelastic effect. Thermal data have been collected using a state-of-the-art photovoltaic effect detector and lower cost, lower thermal resolution microbolometer detectors, during crack propagation induced by both constant amplitude and frequency loading, and by idealized flight cycles. The Euclidean distance calculated between the feature vectors of the initial and current state can be used to indicate the presence of damage. Cracks of the order of 1 mm in length can be detected and tracked, with an increase in the rate of change of the Euclidean distance indicating the onset of critical crack propagation. The differential feature vector method therefore represents a substantial advance in technology for monitoring the initiation and propagation of cracks in structures, both in structural testing and in-service using low-cost sensors.

A Recurrent Neural Network based deep learning model for offline signature verification and recognition system

With the recent advancement in information technology field, the demand to develop a person authentication system through verifying their offline signatures is gradually increasing. This type of system may be used to verify various official documents through verifying the signatures of the concerned persons present in the documents. This article proposes a Recurrent Neural Network (RNN), a deep learning network, based method to verify and recognize offline signatures of different persons. Various structural and directional features have been extracted locally from each signature sample and the generated feature vectors have been studied using two different models of RNN—long-short term memory (LSTM) and bidirectional long–short term memory (BLSTM). The performance of the proposed system has been tested on six widely used public signature databases—GPDS synthetic, GPDS-300, MCYT-75, CEDAR, BHSig260 Hindi, and BHSig260 Bengali. Experiment has also been performed using Convolutional Neural Network (CNN) to have a comparison with RNN based results. Experimental results demonstrate that the proposed RNN based signature verification and recognition system is superior over CNN and also outperforms the existing state-of-the-art results in this regard.

Human face recognition methods based on principle component analysis (PCA), wavelet and support vector machine (SVM) : a comparative study

<p class="JESTECAbstract">Human face Recognition systems are increasingly gaining more importance and can be utilized throughout many applications like video surveillance, Security, human-computer intelligent interaction, etc. this paper presents performance comparison between three feature extraction techniques for an automatic face recognition system. In the first step, we benefit from wavelet Transforms, Principal Component Analysis (PCA) and combining Wavelet with PCA as feature extracting methods. After feature vectors generation, linear and nonlinear Support Vector Machines (SVM) are usually used for implementing the classification or recognition step. These methods are compared on accuracy in an ORL database for face recognition applications including 400 images of 40 people.</p>

Automated diagnosis of breast cancer using parameter optimized kernel extreme learning machine

Early detection and diagnosis is one of the primary steps towards an effective treatment of breast cancer. An automated and reliable computer-aided diagnosis (CAD) model is a suitable alternative to facilitate and ease the diagnosis process. The proposed work focuses on developing an efficacious CAD model that can classify the digital mammograms as normal, benign, or malignant. First, a block-based cross diagonal texture matrix (CDTM) technique is applied to the mammogram ROIs, and then Haralick's features are extracted from each of the ROIs. Then, the kernel principal component analysis (KPCA) technique is used to reduce the dimension of the generated feature vector. Finally, a wrapper-based parameter optimized kernel extreme learning machine (KELM) is proposed wherein the principle of the grasshopper optimization is utilized to determine the optimal values of KELM parameters. Additionally, the proposed wrapper technique is also used to select the important features from the reduced feature vector. The proposed scheme is compared with some of the recently reported CAD models on two standard and publicly available datasets, namely, MIAS and DDSM. From experimental results and analysis, it is noticed that the proposed CAD model yields better results in terms of classification accuracy, and area under curve with a significantly reduced number of features, as compared to that of its counterparts. Additionally, the model is also validated for multi-class classification on the aforementioned datasets reporting an accuracy of 97.49% and 92.61% for MIAS and DDSM datasets, respectively. Further, the proposed model demands minimum computational time than that of its counterparts.