Feature Extraction Approach Research Articles

How to identify patient perception of AI voice robots in the follow-up scenario? A multimodal identity perception method based on deep learning

ObjectivesPost-discharge follow-up stands as a critical component of post-diagnosis management, and the constraints of healthcare resources impede comprehensive manual follow-up. However, patients are less cooperative with AI follow-up calls or may even hang up once AI voice robots are perceived. To improve the effectiveness of follow-up, alternative measures should be taken when patients perceive AI voice robots. Therefore, identifying how patients perceive AI voice robots is crucial. This study aims to construct a multimodal identity perception model based on deep learning to identify how patients perceive AI voice robots. MethodsOur dataset includes 2030 response audio recordings and corresponding texts from patients. We conduct comparative experiments and perform an ablation study. The proposed model employs a transfer learning approach, utilizing BERT and TextCNN for text feature extraction, AST and LSTM for audio feature extraction, and self-attention for feature fusion. ResultsOur model demonstrates superior performance against existing baselines, with a precision of 86.67%, an AUC of 84%, and an accuracy of 94.38%. Additionally, a generalization experimentwas conducted using 144 patients’ response audio recordings and corresponding text data from other departments in the hospital, confirming the model’s robustness and effectiveness. ConclusionOur multimodal identity perception model can identify how patients perceive AI voice robots effectively. Identifying how patients perceive AI not only helps to optimize the follow-up process and improve patient cooperation, but also provides support for the evaluation and optimization of AI voice robots.

Surface feature extraction method for cloud data of aircraft wall panel measurement points

In the cloud, users need to connect to the data server to perform the file transmission via the Internet, and the Server transmits data to many servers. A machine or vehicle that can fly with the assistance of the air is known as an Aircraft. As an alternative to the downward thrust of jet engines, it uses either static lift or an airfoil's dynamic lift to combat gravity's pull. Drawing wall panel measurement points in the model is easy using the Aircraft Wall Panels (AWP) button. Draw wall panels between existing nodes or on the drawing grid using the relevant wall panel specifications. The technique intends to discover and extract information about undesirable defects such as dents, protrusions, or scratches based on local surface attributes gathered from a 3D scanner. Defects from a perfectly smooth surface include indentations and bumps on the surface. An image's features may be extracted by reducing the number of pixels in the picture to a manageable size so that the most exciting sections of the image can be recorded with Surface Feature Extraction (SFE). Some of the problems are the threat of drones and composite materials that do not break easily in oxymoronic. The aircraft's inner structure may have been damaged, although this is impossible to determine. A runway incursion severely threatens aviation safety because of the rise in aircraft movement on the airport surface and other human factors. An electronic moving map of airport runways and taxiways is shown to the pilot through a head-up display in the cockpit's head-down position. A practical feature extraction approach is required to ensure the safety of the airport scene in runway incursion prevention systems. All the drawbacks are rectified by AWP-SFE sensors installed along the runway centerline to detect magnetic signals generated by surface-moving targets, and this information is utilized to compute the target's length. The target length may extract peak features after regularizing the time domain data. Differentiation of target characteristics is used to determine the similarities between distinct targets. The suggested method's signal characteristics are more easily recognized than time domain or frequency domain feature methods. The experimental results show the proposed method AWP-SE to achieve a high-efficiency ratio of 88.2%, activity ratio of 73.3%, Analysis of aircraft in wall plane measurement point of 87.8% and an error rate of 32.3% compared to other methods.

MGSAN: multimodal graph self-attention network for skeleton-based action recognition

Due to the emergence of graph convolutional networks (GCNs), the skeleton-based action recognition has achieved remarkable results. However, the current models for skeleton-based action analysis treat skeleton sequences as a series of graphs, aggregating features of the entire sequence by alternately extracting spatial and temporal features, i.e., using a 2D (spatial features) plus 1D (temporal features) approach for feature extraction. This undoubtedly overlooks the complex spatiotemporal fusion relationships between joints during motion, making it challenging for models to capture the connections between different temporal frames and joints. In this paper, we propose a Multimodal Graph Self-Attention Network (MGSAN), which combines GCNs with self-attention to model the spatiotemporal relationships between skeleton sequences. Firstly, we design graph self-attention (GSA) blocks to capture the intrinsic topology and long-term temporal dependencies between joints. Secondly, we propose a multi-scale spatio-temporal convolutional network for channel-wise topology modeling (CW-TCN) to model short-term smooth temporal information of joint movements. Finally, we propose a multimodal fusion strategy to fuse joint, joint movement, and bone flow, providing the model with a richer set of multimodal features to make better predictions. The proposed MGSAN achieves state-of-the-art performance on three large-scale skeleton-based action recognition datasets, with accuracy of 93.1% on NTU RGB+D 60 cross-subject benchmark, 90.3% on NTU RGB+D 120 cross-subject benchmark, and 97.0% on the NW-UCLA dataset. Code is available at https://github.com/lizaowo/MGSAN.

A new strategy based on multi-source remote sensing data for improving the accuracy of land use/cover change classification.

Land Use/Cover Change (LUCC) plays a crucial role in sustainable land management and regional planning. However, contemporary feature extraction approaches often prove inefficient at capturing critical data features, thereby complicating land cover categorization. In this research, we introduce a new feature extraction algorithm alongside a Segmented and Stratified Principal Component Analysis (SS-PCA) dimensionality reduction method based on correlation grouping. These methods are applied to UAV LiDAR and UAV HSI data collected from land use types (e.g., residential areas, agricultural lands) and specific species (e.g., tree species) in urban, agricultural, and natural environments to reflect the diversity of the study area and to demonstrate the ability of our methods to be applied in different classification scenarios. We utilize LiDAR and HSI data to extract 157 features, including intensity, height, Normalized Digital Surface Model (nDSM), spectral, texture, and index features, to identify the optimal feature subset. Subsequently, the best feature subset is inputted into a random forest classifier to classify the features. Our findings demonstrate that the SS-PCA method successfully enhances downscaled feature bands, reduces hyperspectral data noise, and improves classification accuracy (Overall Accuracy = 91.17%). Additionally, the CFW method effectively screens appropriate features, thereby increasing classification accuracy for LiDAR (Overall Accuracy = 78.10%), HIS (Overall Accuracy = 89.87%), and LiDAR + HIS (Overall Accuracy = 97.17%) data across various areas. Moreover, the integration of LiDAR and HSI data holds promise for significantly improving ground fine classification accuracy while mitigating issues such as the 'salt and pepper noise'. Furthermore, among individual features, the LiDAR intensity feature emerges as critical for enhancing classification accuracy, while among single-class features, the HSI feature proves most influential in improving classification accuracy.

Design of EEG based thought identification system using EMD & deep neural network

Biological communication system for neurological disorder patients is similar to the Brain Computer Interface in a way that it facilitates the connection to the outside world in real time. The interdisciplinary field of Electroencephalogram based message depiction is gaining importance as it assists the paralysed person to communicate. In the proposed method a novel approach of feature extraction is done by Empirical Mode Decomposition on non- stationary & non-linear kind of EEG signal. EMD helps in the effective time frequency analysis by disintegrating the EEG signal in the form of six Intrinsic Mode Functions with help of the frequency components. In all nine features are extracted from the decomposed IMFs so as to predict the states or messages of the patient. The above computed features are then served to the Deep Neural Network to perform the classification. The performance of suggested method is studied through applying it to the acquired database generated by the designed hardware as well as also in real time message depiction. The maximum classification accuracy 97% for the acquired database & 85% in real time are obtained respectively by comparative analysis. The command messages generated from the proposed system helps the person suffering from neurological disorder to establish the communication link with the outside world in an efficient way. Thus, the proposed novel method shows better performance in real time message depiction purpose as related to other existing methods.

Segmentation-based truncated-SVD for effective feature extraction in hyperspectral image classification

ABSTRACT Remote sensing hyperspectral images (HSIs) are rich sources of information about land cover captured across hundreds of narrow, contiguous spectral wavelength bands. However, using the entire original HSI for practical applications can lead to suboptimal classification accuracy. To address this, band reduction techniques, categorized as feature extraction and feature selection methods, are employed to enhance classification results. One commonly used feature extraction approach for HSIs is Principal Component Analysis (PCA). However, PCA may fall short of capturing the local and specific characteristics present in the HSI data. In this paper, we introduce two novel feature extraction methods: Segmented Truncated Singular Value Decomposition (STSVD) and Spectrally Segmented Truncated Singular Value Decomposition (SSTSVD) to improve classification performance. Segmentation is carried out based on highly correlated bands’ segments and spectral bands’ segments within the HSI data. Our study evaluates and compares these newly proposed methods against classical feature extraction methods, including PCA, Incremental PCA, Sparse-PCA, Kernel PCA, Segmented-PCA (SPCA), and Truncated Singular Value Decomposition (TSVD). We perform this analysis on three distinct HSI datasets, namely the Indian Pines HSI, the Pavia University HSI, and the Kennedy Space Center HSI, using per-pixel Support Vector Machine (SVM) and Random Forest (RF) classification. The experimental results demonstrate the superiority of our proposed methods for all three datasets. The best-performing feature extraction methods when classification is performed using an SVM classifier are STSVD3 (89.03%), SSTSVD2 (95.55%), and STSVD3 (97.74%) for the Indian Pines, Pavia University, and Kennedy Space Center datasets, respectively. Similarly, for the RF classifier, the best-performing feature extraction methods are SSTSVD4 (88.98%), SSTSVD3 (96.04%), and SSTSVD4 (96.09%) for Indian Pines, Pavia University, and Kennedy Space Center datasets, respectively.

Sensor-Based Multifaceted Feature Extraction and Ensemble Elastic Net Approach for Assessing Fall Risk in Community-Dwelling Older Adults.

Accurate identification of community-dwelling older adults at high fall risk can facilitate timely intervention and significantly reduce fall incidents. Analyzing gait and balance capabilities via feature extraction and modeling through sensor-based motion data has emerged as a viable approach for fall risk assessment. However, the existing approaches for extracting key features related to fall risk lack inclusiveness, with limited consideration of the non-linear characteristics of sensor signals, such as signal complexity, self-similarity, and local stability. In this study, we developed a multifaceted feature extraction scheme employing diverse feature types, including demographic, descriptive statistical, non-linear, spatiotemporal and spectral features, derived from three-axis accelerometers and gyroscope data. This study is the first attempt to investigate non-linear features related to fall risk in multi-task scenarios from a dynamic system perspective. Based on the extracted multifaceted features, we propose an ensemble elastic net (E-E-N) approach for handling imbalanced data and offering high model interpretability. The E-E-N utilizes bootstrap sampling to construct base classifiers and employs a weighting mechanism to aggregate the base classifiers. We conducted a set of validation experiments using real-world data for comprehensive comparative analysis. The results demonstrate that the E-E-N approach exhibits superior predictive performance on fall risk classification. Our proposed approach offers a cost-effective tool for accurately assessing fall risk and alleviating the burden of continuous health monitoring in the long term.

AI-Based Feature Extraction Approaches for Dual Modalities of Autism Spectrum Disorder Neuroimages

AI-Based Feature Extraction Approaches for Dual Modalities of Autism Spectrum Disorder Neuroimages

Performance Evaluation of Thermography-Based Computer-Aided Diagnostic Systems for Detecting Breast Cancer: An Empirical Study

Among women, breast cancer is one of the most commonly occurring cancers besides skin and cervix cancer. Developing countries are at a higher risk of mortality due to late-stage presentation, inaccessible diagnosis, and treatment. Thermography-based technology, aided with machine learning, for screening/diagnosing breast cancer is non-invasive, cost-wise appropriate, and requires very little equipment in rural areas with limited facilities. In this article, we systematically compare the state-of-the-art feature extraction approaches on a uniform platform, using two common datasets, three feature selection methods, four well-known classifiers, and three cross-validation strategies and analyze the results, for a fair comparison. Also, we evaluated the performance when all the features were combined (Unified Model) on the same platform. Experimental results show that the classification accuracy improves considerably with the use of feature selection methods. Among all the combinations considered, the classification model with Union \(\_\) FeatureSet and mRMR gave the best performance for both datasets. We obtained a feature subset of 26 and 34 features (from Union \(\_\) FeatureSet) with a combination of mRMR and SVM, which are relevant, non-redundant, and distinguish normal and abnormal thermal patterns with the accuracy of 95.73% on the DMR-IR dataset and 92.533% on the RGC-IR dataset.

Ensemble feature selection and tabular data augmentation with generative adversarial networks to enhance cutaneous melanoma identification and interpretability

BackgroundCutaneous melanoma is the most aggressive form of skin cancer, responsible for most skin cancer-related deaths. Recent advances in artificial intelligence, jointly with the availability of public dermoscopy image datasets, have allowed to assist dermatologists in melanoma identification. While image feature extraction holds potential for melanoma detection, it often leads to high-dimensional data. Furthermore, most image datasets present the class imbalance problem, where a few classes have numerous samples, whereas others are under-represented.MethodsIn this paper, we propose to combine ensemble feature selection (FS) methods and data augmentation with the conditional tabular generative adversarial networks (CTGAN) to enhance melanoma identification in imbalanced datasets. We employed dermoscopy images from two public datasets, PH2 and Derm7pt, which contain melanoma and not-melanoma lesions. To capture intrinsic information from skin lesions, we conduct two feature extraction (FE) approaches, including handcrafted and embedding features. For the former, color, geometric and first-, second-, and higher-order texture features were extracted, whereas for the latter, embeddings were obtained using ResNet-based models. To alleviate the high-dimensionality in the FE, ensemble FS with filter methods were used and evaluated. For data augmentation, we conducted a progressive analysis of the imbalance ratio (IR), related to the amount of synthetic samples created, and evaluated the impact on the predictive results. To gain interpretability on predictive models, we used SHAP, bootstrap resampling statistical tests and UMAP visualizations.ResultsThe combination of ensemble FS, CTGAN, and linear models achieved the best predictive results, achieving AUCROC values of 87% (with support vector machine and IR=0.9) and 76% (with LASSO and IR=1.0) for the PH2 and Derm7pt, respectively. We also identified that melanoma lesions were mainly characterized by features related to color, while not-melanoma lesions were characterized by texture features.ConclusionsOur results demonstrate the effectiveness of ensemble FS and synthetic data in the development of models that accurately identify melanoma. This research advances skin lesion analysis, contributing to both melanoma detection and the interpretation of main features for its identification.

Enhancing induction machine fault detection through machine learning: Time and frequency analysis of vibration signals

The integration of machine learning algorithms into fault diagnosis is regarded as an advanced and effective method for detecting electrical system faults. Vibration signals are identified as valuable and easily accessible data for fault identification in electrical machines. In this study, experiments were conducted to assemble a dataset consisting of 525 × 3 instances, each containing 200 three-axis vibration signal measurements. These measurements were obtained from a laboratory test bench equipped with a customized winding three-phase induction machine. Instead of directly analyzing the vibration signals, a feature extraction approach was employed, calculating a comprehensive set of statistical, harmonic, and impulsive features from the time domain (e.g., ShapeFactor, SINAD, ClearanceFactor, ImpulseFactor, CrestFactor, Kurtosis, Skewness, THD, Std, RMS, Mean, PeakValue, SNR), along with spectral features from the frequency domain (e.g., Peak Frequencies, Amplitudes, BandPower). To improve model efficiency, Analysis of Variance (ANOVA) was applied to select only the most significant features, with F-statistics used to rank feature importance. Features with higher F-statistics were prioritized for their ability to explain more variance in motor fault classification. This selective approach reduced model complexity, helping to minimize overfitting and focus on features that had a substantial impact on predictive accuracy. By concentrating on these high-impact features, a balance between simplicity and performance was achieved. These selected features were then used to train five machine learning classifiers: Ensemble (Bagged Trees), Quadratic Support Vector Machine, Weighted K-Nearest Neighbors, Efficient Logistic Regression, and Neural Networks. The performance of these classifiers was evaluated using various metrics, with validation accuracy rates ranging from 78.3% to 98.8%. A comparative study with similar works in the literature was conducted, demonstrating that high performances were obtained with the proposed approach while maintaining computational efficiency.

An Investigation into the Utilisation of CNN with LSTM for Video Deepfake Detection

Video deepfake detection has emerged as a critical field within the broader domain of digital technologies driven by the rapid proliferation of AI-generated media and the increasing threat of its misuse for deception and misinformation. The integration of Convolutional Neural Network (CNN) with Long Short-Term Memory (LSTM) has proven to be a promising approach for improving video deepfake detection, achieving near-perfect accuracy. CNNs enable the effective extraction of spatial features from video frames, such as facial textures and lighting, while LSTM analyses temporal patterns, detecting inconsistencies over time. This hybrid model enhances the ability to detect deepfakes by combining spatial and temporal analysis. However, the existing research lacks systematic evaluations that comprehensively assess their effectiveness and optimal configurations. Therefore, this paper provides a comprehensive review of video deepfake detection techniques utilising hybrid CNN-LSTM models. It systematically investigates state-of-the-art techniques, highlighting common feature extraction approaches and widely used datasets for training and testing. This paper also evaluates model performance across different datasets, identifies key factors influencing detection accuracy, and explores how CNN-LSTM models can be optimised. It also compares CNN-LSTM models with non-LSTM approaches, addresses implementation challenges, and proposes solutions for them. Lastly, open issues and future research directions of video deepfake detection using CNN-LSTM will be discussed. This paper provides valuable insights for researchers and cyber security professionals by reviewing CNN-LSTM models for video deepfake detection contributing to the advancement of robust and effective deepfake detection systems.

Attention Induced Dual Convolutional-Capsule Network (AIDC-CN): A deep learning framework for motor imagery classification

In recent times, Electroencephalography (EEG)-based motor imagery (MI) decoding has garnered significant attention due to its extensive applicability in healthcare, including areas such as assistive robotics and rehabilitation engineering. Nevertheless, the decoding of EEG signals presents considerable challenges owing to their inherent complexity, non-stationary characteristics, and low signal-to-noise ratio. Notably, deep learning-based classifiers have emerged as a prominent focus for addressing the EEG signal decoding process. This study introduces a novel deep learning classifier named the Attention Induced Dual Convolutional-Capsule Network (AIDC-CN) with the specific aim of accurately categorizing various motor imagination class labels. To enhance the classifier’s performance, a dual feature extraction approach leveraging spectrogram and brain connectivity networks has been employed, diversifying the feature set in the classification task. The main highlights of the proposed AIDC-CN classifier includes the introduction of a dual convolution layer to handle the brain connectivity and spectrogram features, addition of a novel self-attention module (SAM) to accentuate the relevant parts of the convolved spectrogram features, introduction of a new cross-attention module (CAM) to refine the outputs obtained from the dual convolution layers and incorporation of a Gaussian Error Linear Unit (GELU) based dynamic routing algorithm to strengthen the coupling among the primary and secondary capsule layers. Performance analysis undertaken on four public data sets depict the superior performance of the proposed model with respect to the state-of-the-art techniques. The code for this model is available at https://github.com/RiteshSurChowdhury/AIDC-CN.

MFCA-MICNN: a convolutional neural network with multiscale fast channel attention and multibranch irregular convolution for noise removal in dMRI

Diffusion magnetic resonance imaging (dMRI) currently stands as the foremost noninvasive method for quantifying brain tissue microstructure and reconstructing white matter fiber pathways. However, the inherent free diffusion motion of water molecules in dMRI results in signal decay, diminishing the signal-to-noise ratio (SNR) and adversely affecting the accuracy and precision of microstructural data. In response to this challenge, we propose a novel method known as the Multiscale Fast Attention-Multibranch Irregular Convolutional Neural Network for dMRI image denoising. In this work, we introduce Multiscale Fast Channel Attention, a novel approach for efficient multiscale feature extraction with attention weight computation across feature channels. This enhances the model's capability to capture complex features and improves overall performance. Furthermore, we propose a multi-branch irregular convolutional architecture that effectively disrupts spatial noise correlation and captures noise features, thereby further enhancing the denoising performance of the model. Lastly, we design a novel loss function, which ensures excellent performance in both edge and flat regions. Experimental results demonstrate that the proposed method outperforms other state-of-the-art deep learning denoising methods in both quantitative and qualitative aspects for dMRI image denoising with fewer parameters and faster operational speed.

Comparative analysis of features extraction protocols for LC-HRMS untargeted metabolomics in mountain cheese ‘identitation’

This study presents a comprehensive metabolomic analysis of Parmigiano Reggiano samples to differentiate between those designated as Mountain Quality Certification (QC) and conventional Protected Designation of Origin (PDO). Despite following the same production protocol, these cheese varieties differ in the cows’ feeding regimes and milk stable locations, with mountain-certified samples adhering to specific requirements regarding milk origin and feed composition. An untargeted approach with Liquid Chromatography-High Resolution Mass Spectrometry (LC-HRMS) was proposed to characterize the cheese metabolome. High-resolution LC-MS data can generate gigabyte-sized files, making data compression essential for manageable multivariate analysis and noise reduction. This study employs the Region of Interest-Multivariate Curve Resolution (ROI-MCR) protocol to achieve effective data compression and chromatographic resolution, thereby extracting the most informative features. This method was compared with a classical approach for feature extraction from chromatographic data, namely Compound Discoverer (CD) software. The features extracted by both methods were analysed through Principal Component Analysis (PCA) and ASCA (ANOVA Simultaneous Component Analysis). The comparison of ROI-MCR and CD approaches demonstrated that while both methods yielded similar overall conclusions, ROI-MCR provided a more streamlined and manageable dataset, facilitating easier interpretation of the metabolic differences. Both approaches indicated that amino acids, fatty acids, and bacterial activity-related compounds played significant roles in distinguishing between the two sample types.

Research on fiber OTDR signal recognition and classification based on the ViT model

In the field of event recognition for phase-sensitive optical time-domain reflectometry (-OTDR), convolutional neural networks (CNNs) have been the mainstream tool. However, Transformer models, with their self-attention mechanism, have provided a new perspective for image recognition tasks. This paper proposes a -OTDR event recognition method based on the Transformer model and experimentally demonstrates its significant advantages over CNNs. The method utilizes an attention-based feature extraction approach to recognize distributed fiber optic sensing signals in a phase-sensitive optical time-domain reflectometer (-OTDR). In perimeter defense applications, various interferences and noise severely affect the recognition of -OTDR sensing signals, leading to false alarms, making accurate signal recognition both challenging and necessary. Extracting signal features and using machine learning classification models to recognize signals has been a research hotspot in recent years. The effectiveness of feature extraction is crucial to classification performance. This project introduces a temporal-spatial dual attention mechanism into the feature extraction of fiber optic sensing signals to improve the accuracy and robustness of signal recognition.

Enhanced detection of diabetes mellitus using novel ensemble feature engineering approach and machine learning model

Diabetes is a persistent health condition led by insufficient use or inappropriate use of insulin in the body. If left undetected, it can lead to further complications involving organ damage such as heart, lungs, and eyes. Timely detection of diabetes helps obtain the right medication, diet, and exercise plan to lead a healthy life. ML approach has been utilized to obtain rapid and reliable diabetes detection, however, existing approaches suffer from the use of limited datasets, lack of generalizability, and lower accuracy. This study proposes a novel feature extraction approach to overcome these limitations by using an ensemble of convolutional neural network (CNN) and long short-term memory (LSTM) models. Multiple datasets are combined to make a larger dataset for experiments and multiple features are utilized for investigating the efficacy of the proposed approach. Features from the extra tree classifier, CNN, and LSTM are also considered for comparison. Experimental results reveal the superb performance of CNN-LSTM-based features with random forest model obtaining a 0.99 accuracy score. This performance is further validated by comparison with existing approaches and k-fold cross-validation which shows the proposed approach provides robust results.

Text mining approach for feature extraction and cartilage disease grade classification using knee MRI radiology reports

MRI radiology reporting processes can be improved by exploiting structured and semantically labelled data that can be fed to artificial intelligence (AI) tools. AI-based tools assisting radiology reporting can help to automatically individuate cartilage grading in textual magnetic resonance imaging (MRI) reports, thus supporting clinicians' decisions regarding medical imaging utilisation, diagnosis and treatment. In this study, we extracted information (clinical findings, observations, anatomical regions, etc.) and classified knee cartilage degradation from medical reports utilising transfer-learning techniques applied to the Bidirectional Encoder Representations from Transformers (BERT) model and its variants, pre-trained on an Italian-language corpus. To realise this objective, we used a dataset of 750 MRI knee reports written by three radiologists who contributed to a manual annotation process to perform text classification (TC) and named entity recognition (NER) tasks. The dataset was obtained from an internal database of the IRCCS SYNLAB SDN. Seventy percent of the dataset was used for training, 10% was used for validation and 20% was used for testing. The best-performing configurations for NER and TC tasks were based on the pre-trained BERT model. The macro F1-scores obtained with the NER and TC models are 0.89 and 0.81, respectively. The accuracies calculated on the test set for both tasks are 0.96 and 0.99, respectively.

A rolling bearing failure feature extraction approach based on IBWO-VME-MCKD

A rolling bearing failure feature extraction approach based on IBWO-VME-MCKD

Comparative analysis of speaker identification performance using deep learning, machine learning, and novel subspace classifiers with multiple feature extraction techniques

Speaker identification is vital in various application domains, such as automation, security, and enhancing user experience. In the literature, convolutional neural network (CNN) or recurrent neural network (RNN) classifiers are generally used due to the one-dimensional time series of speech signals. However, new approaches using subspace classifiers are also crucial in speaker identification. In this study, in addition to the newly developed subspace classifiers for speaker identification, traditional classification algorithms, and various hybrid algorithms are analyzed in terms of performance. Stacked Features-Common Vector Approach (SF-CVA) and Hybrid CVA-Fisher Linear Discriminant Analysis (HCF) subspace classifiers are used for speaker identification for the first time in the literature. In addition, CVA is evaluated for the first time for speaker identification using hybrid deep learning algorithms. The study includes Recurrent Neural Network-Long Short-Term Memory (RNN-LSTM), i-vector + Probabilistic Linear Discriminant Analysis (i-vector+PLDA), Time Delayed Neural Network (TDNN), AutoEncoder+Softmax (AE+Softmax), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Common Vector Approach (CVA), SF-CVA, HCF, and Alexnet classifiers for speaker identification. This study uses MNIST, TIMIT and Voxceleb1 databases for clean and noisy speech signals. Six different feature structures are tested in the study. The six different feature extraction approaches consist of Mel Frequency Cepstral Coefficients (MFCC)+Pitch, Gammatone Filter Bank Cepstral Coefficients (GTCC)+Pitch, MFCC+GTCC+Pitch+seven spectral features, spectrograms,i-vectors, and Alexnet feature vectors. High accuracy rates were obtained, especially in tests using SF-CVA. RNN-LSTM, i-vector+KNN, AE+Softmax, TDNN, and i-vector+HCF classifiers also gave high test accuracy rates.