Search

Vibration-based signal processing is the most popular and effective approach for fault diagnosis of bearing. In this paper, time-frequency domain analysis, i.e. empirical mode decomposition (EMD) was applied to the raw vibration signal. Intrinsic mode function (IMF) containing the characteristics of vibration data was analysed to obtain 90 statistical features. Two feature reduction algorithms, namely principal components analysis (PCA) and binary particle swarm optimiser (BPSO) were applied individually for feature reduction. The reduced feature subsets were 12 and 35 for PCA and BPSO, respectively. K-Nearest Neighbours (K-NN) was used as an intelligent method for fault diagnosis. K-NN was applied to the entire feature set and individually on the selected feature subset of PCA and BPSO. The reduced feature subset with PCA performed the finest in all the measurements taken. For BPSO, although it effectively reduced the feature dimension and classification time, the testing accuracy was slightly lower. Comparing the output accuracy of the K-NN classifier for the selected methods demonstrated the effectiveness of PCA and BPSO as efficacious feature reduction techniques.

Dermoscopic images carry rich information to identify the malignancy in patients at initial stage. Research initiatives in the domain of content-based image classification can be instrumental in identifying fatal diseases like skin cancer by exploring the dermoscopic image database. This paper has carried out feature dimension reduction for representation of significant content-based image descriptors to the classifiers. The approach has resulted in designing an early fusion based classification model with reduced computational overhead to enhance accuracy of malignancy detection at its inception.

Tracking accuracy and frame rate are the two most important indexes of satellite video tracking. Now, the research of close-range tracking algorithm is gradually developing from short-term to long-term, and has been widely used. This paper refers to the idea of long-term tracking strategy and introduces it into satellite video target tracking. With a variety of features as precision guarantee, the redetection function is added to the tracker to ensure the robust tracking. At the same time, the feature dimension is properly reduced. The validity of the algorithm is verified by the video sequence obtained from Skybox-1 satellite. The results show that the tracking effect is good, and the performance of the dimension-reduced version has no significant decrease, and the frame rate is improved by 37.8%. It shows the high feasibility and wide application prospect of long-term tracking strategy in space-based video tracking.

In order to better mine and analyze effective information from massive electric power big data and improve the short-term power prediction accuracy of photovoltaic power generation data, this paper proposes a short-term power prediction method based on improved kernel principal component analysis and back propagation neural network optimized by particle swarm optimization (IKPCA-PSO-BPNN). The IKPCA method is used to reduce the feature dimension, and the processed data is trained and predicted in the PSO-BPNN model, which improves the model performance and prediction accuracy.

Handwriting recognition is a challenging task. In this paper, characters are first extracted from the English, Arabic, and Chinese handwriting dataset. The handwritten content recognition process comprises four stages, namely information preprocessing, segmentation, feature extraction, and dimension. The character segmentation algorithm is based on forward and backward classification. The present work implements the feature dimension sequence labeling method (FEDSEL) and semantic role labeling (SRL) for handwritten recognition. The dataset used is the HCAR database. The accuracy of the proposed method is 95.77%.

Hyperspectral image classification has so many applications in the area of remote sensing. In recent years, deep learning has been accepted as a powerful tool for feature extraction and ensuring better classification accuracies. In this paper, model for HSI classification is created by implementing open set domain adaptation and generative adversarial networks (GAN). Open set domain adaptation is a type of domain adaptation where target has more classes which are not present in the source distribution. Huge dimension of hyperspectral image needs to be reduced for an efficient classification. In this work, we analysed the effect of dimensionality reduction for open set domain adaptation for hyperspectral image classification by using dynamic mode decomposition (DMD) technique. Experimental results show that 20% of the total available bands of Salinas and 30% of the bands of PaviaU dataset are the highest achievable reduction in feature dimension that results in almost same classification accuracy.

In this paper, two types of features ‘angle’ feature and \(Finger\_{Tips}\) distance feature extraction methods for gestures of finger recognition are proposed. The entire image is segmented into several spatial modules and the task of feature extraction is carried out on finger of the hand images. Application of this method is extended to medical systems, sign languages for hearing-impaired people, crisis management and disaster relief, entertainment and human- -robot interaction. This method is tested on medial axis transformation (MAT) image and it does not require any gloves for recognition. This feature extraction algorithm has an advantage of very low feature dimension.

The 3-dimensional (3D) super-resolution (SR) for medical volumetric data is confirmed to provide better visual results compared to conventional 2-dimensional processing. Then, considering practical applications, we propose a novel convolutional neural network for a 3D SR. Moreover, to reduce model parameters and achieve high precision, we reformulating the standard convolution layer by shrinking the input feature dimension before mapping and expanding back afterward using pointwise convolution. Thus, we utilize channel attention to optimize the network for optimum performance. Although the proposed architecture can be embedded into any 3D model, it performs better while combined with the state-of-the-art model, the 3D residual dense network (RDN). Furthermore, we demonstrate that our proposed method outperforms the state-of-the-art methods in terms of accuracy with only approximately 1/8 parameters and 3/5 times of the standard model, 3D RDN. Consequently, our network is quite suitable in practical situations, such as small training samples, short processing time, and being embedded into chips.

“Large Dimension” of features derived from protein sequences is a major problem in protein–protein interaction (PPI) prediction. Thus, reduction of the feature dimension may increase the classification accuracy. In this paper, particle swarm optimization (PSO) and principal component analysis (PCA) have been used for dimensionality reduction of PPI sequence features. The performance of the algorithm has been assessed using the intraspecies E coli protein–protein interaction database, containing an equal number of positive and negative interacting pairs. Standard sequence-based features such as amino acid composition (AAC), dipeptide composition (Dipep), and conjoint triad composition (CTD) are extracted. From the results, it is seen that the PSO-based dimensionality reduction method provides steady and better performance in terms of accuracy when applied to the features.

This paper compares the classification ability of a few efficient classifiers in recognizing human speech emotions in terms of accuracy, computation time, and feature dimension. Both the static and dynamic mel-frequency cepstral coefficients (MFCC) are derived in the wavelet domain and are combined to form a suitable identification system model. Three popular neural network (NN) models such as the Gaussian mixture model (GMM), radial basis function network (RBFN), and the probabilistic neural network (PNN) have been put to test the reliability of these derived feature sets. The PNN classifier has shown to outperform both the RBFN and the GMM with low feature dimension, whereas the GMM shows an improved result for large feature dimensions. The combination of wavelet-based MFCCs and their dynamics remains more discriminative in classifying speech emotions as compared to either the MFCCs or wavelets acted alone.

High-dimensional features have great advantages for the analysis and identification of radio fingerprint features. In order to enhance the classification and recognition capability of frequency hopping stations, it is usually necessary to increase the feature type and dimension to further improve the classification accuracy of the classifier. However, with the increase of the feature types and the dimensions, a large number of irrelevant and redundant features will be introduced, which leads to the increased classification time and the low classification accuracy. In order to reduce the feature dimension and remove redundant features, a FCBF feature selection algorithm based on normalized mutual information was proposed, named FCBF-NMI. The algorithm uses normalized mutual information instead of symmetric uncertainty as the correlation evaluation standard of FCBF algorithm, and analyzes the correlation between features and categories, deletes irrelevant and redundant features, and finally obtains optimal feature subset. Experimental results show that, FCBF-NMI can obtain the reasonable optimal features, on the base of guaranteeing the correct classification rate, the computing time can be reduced, and the effectiveness of feature recognition and the generalization ability of classification algorithms can be improved as well.

The traditional method of fault diagnosis is essentially looking for the optimal combination of feature extractor and classifier. In this process, it is necessary to manually extract the expert knowledge of features and related fields, which greatly limits the versatility and generalization of the algorithm. The convolutional neural network has the characteristics of “end-to-end”, which can directly perform the whole process of feature extraction, feature dimension reduction and classifier classification on the original signal through a neural network. The traditional convolutional neural network model has the same convolution kernel shape per convolutional layer, and the feature extraction ability is relatively limited. However, the fault signals of equipment or components are often complex and variable, and data features are difficult to mine. In view of the above problems, this paper proposes a fault diagnosis method based on multi-scale convolutional neural network. Based on the traditional convolutional neural network, the diversity of convolutional layer convolution kernels is increased. Finally, the feature data extracted by each scale convolution kernel is merged. The experiment proposed in this paper has a high fault recognition rate by experimenting with the bearing fault public data set.

In this paper, we propose a new framework for affective video content analysis by using two compact audio-visual features. In the proposed framework, the eGeMAPS is first calculated as global audio feature and then the key frames of optical flow images are fed to VGG19 network for implementing the transfer learning and visual feature extraction. Finally for model learning, the logistic regression is employed for affective video content classification. In the experiments, we perform the evaluations of audio and visual features on the dataset of Affective Impact of Movies Task 2015 (AIMT15), and compare our results with those of competition teams participated in AIMT15. The comparison results show that the proposed framework can achieve the comparable classification result with the first place of AIMT15 with a total feature dimension of 344, which is only about one thousandth of feature dimensions used in the first place of AIMT15.

In fault level identification, there are ordinal structures between different levels, and some features have the monotonic dependency with decision. In this paper, we propose a feature selection algorithm for fault level identification based on ordinal classification. First, we design a new feature evaluation function to evaluate the quality of features based on ordinal rough set. Second, combining with the search strategy of genetic algorithm (GA), an improved feature selection algorithm is proposed. Finally, the proposed feature selection algorithm is employed to crack level identification. Experimental results show that the proposed algorithm not only can reduce the feature dimension but also improve the accuracy of identification.

This paper hopes to predict the corrosion potential of steel by establishing a model of the relationship between the corrosion potential of steel in the marine environment and its component content. However, the number of experimental samples in this paper is very small, only 14, while the experimental sample has a high feature dimension, 22 dimensions. Firstly, by increasing the factor level of seawater environment, the experimental samples are expanded by orthogonal experiment method, and then the dimensionality of the samples is reduced by using locality preserving projections, which improves the problem of high-dimensional small samples of data. Finally, the extreme learning machine algorithm and the particle swarm-extreme learning machine optimization algorithm are used to build the model respectively. The results show that the prediction accuracy of the PSO-ELM model is better than that of the ELM model.

For the high dimension of fingerprint feature set in the process of specific emitter identification (SEI), feature selection method is utilized to reduce the feature dimension and improve individual recognition rate. This paper adopted the filter feature selection in four ways: MIFS, mRMR, CMIM, and JMIM fingerprint feature set of high-dimensional feature selection and combined with PCA dimensionality reduction algorithm to minimize the feature dimension. The simulation results show that feature selection is feasible in individual recognition of the radiation source and can be effectively combined with dimension reduction algorithm.

Handwritten digit recognition is a well-studied pattern recognition problem. Most of the techniques, reported in the literature, have been concentrated on designing several feature vectors which represent the digits in a better way. But, at times, such attempt not only increases the dimensionality of extracted feature vector but also suffers from having irrelevant and/or redundant features. To address this, in the present work, a recently introduced Particle Swarm Optimization (PSO) based feature selection method has been applied with suitable modifications. For the course of this experiment, we have confined ourselves to a newly employed feature vector for handwritten digit recognition, namely DAISY feature descriptor. The proposed feature selection method is tested on three handwritten digit databases written in Bangla, Devanagari and Roman scripts. The experimental results show that significant amount of feature dimension is reduced without compromising on recognition accuracy. Comparison of the present feature selection method with two of its ancestors also reveals that the present method outperforms the others.

With the rapid development of the Internet, the high dimensional text data has increased rapidly. How to build an efficient and extensible text classification algorithm has become a hot topic in the field of data mining. Aiming at the problems of high feature dimension, sparse data and long computation time in traditional SVM classification algorithm based on TF-IDF (Term Frequency-Inverse Document Frequency), we propose a novel hybrid system for Chinese text classification: CSVM, which is independent of the hand-designed features and domain knowledge. Firstly, the encoding words are done by constructing a text vocabulary of size m for the input language, and then quantize each word using 1-of-m encoding. Secondly, we exploit the CNN (Convolutional Neural Network) to extract the morphological features of character vectors from each word, and then through large scale text material training the semantic feature of each word vectors are be obtained the semantic feature of each word vectors. Finally, the text classification is carried out with the SVM multiple classifier. Testing on a text dataset with 10 categories, the experimental results show that the CSVM algorithm is more effective than other traditional Chinese text classification algorithm.

The rapid advances in mobile and networking technologies results in usage of mobiles for critical applications like m-commerce, m-payments etc. Even though mobile based services offer many benefits, authenticating the user logging into the system is a big challenge. To mitigate this concern, secure mobile applications based on user online signature verification (OSV) has been proposed. Unfortunately, these models would intensify the substantial computational overhead on thin and resource-constrained mobile devices. This summarizes for a critical need of OSV models which are computationally efficient and achieves higher classification accuracy. Recently, several OSV models have been defined in the literature. However, these models are not computationally effective for resource-constrained mobile devices, because the proposed verification models ought to require not only higher feature dimension but also heavy weight writer specific parameter fixation logic. In this manuscript, we propose an efficient and light weight OSV model for resource-constrained mobile devices. Our approach employs dimensionality reduction based on DBSCAN clustering technique and user specific parameter selection. Thorough experimental analysis are conducted on benchmarking online signature datasets MCYT-100 (DB1) and MCYT-330 (DB2) datasets which confirms the efficiency of proposed model with latest OSV models.

The proposed air insulation prediction model is established based on support vector machine. In this chapter, the theoretical basis and implementation procedure of the prediction method are introduced in detail, mainly including the fundamental of statistical learning theory and SVM, the parameter optimization methods, the feature dimension reduction methods, the sample selection method, the error analysis method and the implementation process of the prediction model.

Hyperspectral images (HSIs) cover a wide range of spectral bands in the electromagnetic spectrum with a very finite interval, and with high spectral resolution of data. The main challenges encountered with HSIs are those associated with their large dimensions. To overcome these challenges we need a healthy classification technique, and we need to be able to extract required features. This chapter analyzes the effect of dimensionality reduction on vectorized convolution neural networks (VCNNs) for HSI classification. A VCNN is a recently introduced deep-learning architecture for HSI classification. To analyze the effect of dimensionality reduction (DR) on VCNN, the network is trained with dimensionally reduced hyperspectral data. The network is tuned in accordance with the learning rate and number of iterations. The effect of a VCNN is analyzed by computing overall accuracy, classification accuracy, and the total number of trainable parameters required before and after DR. The reduction technique used is dynamic mode decomposition (DMD), which is capable of selecting most informative bands using the concept of eigenvalues. Through this DR technique for HSI classification using a VCNN, comparable classification accuracy is obtained using the reduced feature dimension and a lesser number of VCNN trainable parameters.

Massive Online Open Courses (MOOCs) are popular these days as they provide a new way for learners. However, MOOC platforms are incapable of detecting learners’ emotions and cannot provide personalized assistance which is a key factor toward learners’ learning performance. In this paper, we propose SmartPlayer which is a framework of emotion recognition taking learners’ appearance and video interaction events into consideration and predicts learners’ emotion. Appearance features and interaction features are extracted from raw data. We applied Support Vector Machine algorithm to train a multi-class classifier. Data are collected, and the accuracy of the classifier reaches 71.89%. We also applied Principal Component Analysis method to reduce feature dimension and trained a new classifier with accuracy decreasing to 71.05%.

Nowadays, feature selection is considered as a de facto standard in the field of pattern recognition where high-dimensional feature attributes are used. The main purpose of any feature selection algorithm is to reduce the dimensionality of the input feature vector while improving the classification ability. Here, a Memetic Algorithm (MA)-based wrapper–filter feature selection method is applied for the recognition of handwritten word images in segmentation-free approach. In this context, two state-of-the-art feature vectors describing texture and shape of the word images, respectively, are considered for feature dimension reduction. Experimentation is conducted on handwritten Bangla word samples comprising 50 popular city names of West Bengal, a state of India. Final results confirm that for the said recognition problem, subset of features selected by MA produces increased recognition accuracy than the individual feature vector or their combination when applied entirely.

The rapidly growing big data generated by connected sensors, devices, the web and social network platforms, etc., have stimulated the advancement of data science, which holds tremendous potential for problem solving in various domains. How to properly utilize the data in model building to obtain accurate analytics and knowledge discovery is a topic of great importance in data mining, and wherefore two issues arise: how to select a critical subset of features and how to select a critical subset of data points for sampling. This paper presents ongoing research that suggests: 1. the critical feature dimension problem is theoretically intractable, but simple heuristic methods may well be sufficient for practical purposes; 2. there are big data analytic problems where evidence suggest that the success of data mining depends more on the critical feature dimension than the specific features selected, thus a random selection of the features based on the dataset’s critical feature dimension will prove sufficient; and 3. The problem of critical sampling has the same intractable complexity as critical feature dimension, but again simple heuristic methods may well be practicable in most applications; experimental results with several versions of the heuristic method are presented and discussed. Finally, a set of metrics for data quality is proposed based on the concepts of critical features and critical sampling.

Brain–Computer Interfacing helps in creation of a communication pathway between brain and external device such that the biological modality of performing the task could be bypassed. This necessitates fast and reliable decoding of brain signals which mandate feature selection to play a crucial role. The literature discloses the improvement in performance of left/right motor imagery signal classification with many-objective feature selection where several classification performance metrics have been maximized for obtaining a good quality feature set. This work analyses the classification performance by varying the feature dimension and number of objectives. A recent many-objective optimization coupled with objective reduction algorithm viz. \(\alpha \)-DEMO has been used for modeling the feature selection as an optimization problem with six objectives. The results obtained in this work have been statistically validated by Friedman Test.

Insulator is the most common equipment in power system, and the recognition of insulator in infrared image is affected seriously under unconstrained condition with angle change. Due to the insulators’ angle diversification and the manually-designed features’ limitation, the existing algorithms have a problem of low accuracy in classification; therefore how to obtain the rotation invariant representations to cope with these adverse conditions is a very important problem. Deep Convolutional Neural Networks (DCNNs) have established a remarkable performance in image classification. The deep features obtained at the top fully-connected layer of the DCNNs (FC-features) exhibit rich global semantic information and are extremely effective in image classification. In this paper, we present a rotation invariant representation generation method named PFE-FDS (Parallel Feature Extraction and Feature Dimension Selection) for infrared insulator recognition which is based on parallel DCNNs FC-features extraction as well as feature sorting and dimension selection based on mutual information to eliminate redundancy. Then the SVM (Support Vector Machine) classifier is trained on our standard infrared insulator dataset for classification. The experimental accuracy shows that the PFE-FDS can improve the accuracy of multi-angle object recognition.

The systems based on image processing for vehicle type recognition is becoming more and fiercer. It plays an important role in traffic safety. In order to improve the problems that traditional Convolutional Neural Network has low accuracy of feature extraction from the low-resolution image, a novel model based on Deep Convolutional Neural Network (DCNN) was proposed. In this paper, our work mainly contains two aspects both extraction of feature dimension and recognition of vehicle image. Firstly, the learning way was introduced, and the raw image of vehicle subsampled with several different sizes was operated with the filter corresponding each channel in a way of convolution to extract the feature dimension of image. Secondly, the features dimension obtained from every channel were merged by a full connected layer. Eventually, features used to recognize the type of vehicle is got. The experiment shows that the architecture of DCNN model has a efficient performance on the recognition of vehicle image. Compared with the traditional algorithm of CNN, the results of experiment show that the mode of DCNN can achieve 97.6% accuracy and a higher precision is got.

Feature selection plays a key role to reduce the high-dimensionality of feature space in machine learning applications by discarding irrelevant and redundant features with the aim of obtaining a subset of features that accurately describe a given problem with a minimum or no degradation of performance. In this paper, a Memetic Algorithm (MA) based Wrapper-filter feature selection framework is proposed for the recognition of handwritten Bangla city names. For evaluating the MA framework, a recently published feature extraction technique, reported in [1], is used for the said pattern recognition problem. Experimentation is conducted on an in-house dataset of 6000 words written in Bangla script. Here, 40 most popular city names of West Bengal, a state in India, have been considered to prepare the dataset. Proposed technique not only reduces the feature dimension, but also enhances the performance of the word recognition technique significantly.

In this paper, an automatic shoeprint retrieval method used in forensic science is proposed. The proposed method extracts shoeprint features using recently reported descriptor called neural code. The first step of feature extraction is rotation compensation. Then, shoeprint image is divided into top region and bottom region, and two neural codes for both regions are obtained. Afterwards, a matching score between test image and reference image is calculated. The matching score is a weighted sum of cosine similarities of both regions’ neural codes. Experimental results show that our method outperforms other methods on a large-scale database captured by commercial shoeprint scanners. By using PCA, the performance can be improved while the feature dimension is reduced dramatically. To our knowledge, this is the first study using the database collected by commercial shoeprint scanners, and our method obtained a cumulative match score of 88.7% at top 10.

Dimension reduction is useful approach in data analysis application. In this paper, research is done to test whether the concept of Dimension reduction can be applied to improve writer verification process results. Two approaches have been chosen to be compared which are Features selection and Feature transformation, where the comparison is on the way of reducing the dimension of writer handwritten data. Both approaches have slightly difference results in reducing the data and classification accuracy. The objective of this paper is to observe the differences between both approaches according to the classification accuracy results, by using some classification techniques.