Feature Extraction Method Research Articles

Automated Hyperspectral Ore–Waste Discrimination for a Gold Mine: Comparative Study of Data-Driven and Knowledge-Based Approaches in Laboratory and Field Environments

Hyperspectral imaging has been increasingly used in mining for detailed mineral characterization and enhanced ore–waste discrimination, which is essential for optimizing resource extraction. However, the full deployment of this technology still faces challenges due to the variability of field conditions and the spectral complexity inherent in real-world mining environments. In this study, we compare the performance of two approaches for ore–waste discrimination in both laboratory and actual mine site conditions: (i) a data-driven feature extraction (FE) method and (ii) a knowledge-based mineral mapping method. Rock samples, including ore and waste from an open-pit gold mine, were obtained and scanned using a hyperspectral imaging system under laboratory conditions. The FE method, which quantifies the frequency absorption peaks at different wavelengths for a given rock sample, was used to train three discriminative models using the random forest classifier (RFC), support vector classification (SVC), and K-nearest neighbor classifier (KNNC) algorithms, with RFC achieving the highest performance with an F1-score of 0.95 for the laboratory data. The mineral mapping method, which quantifies the presence of pyrite, calcite, and potassium feldspar based on prior geochemical analysis, yielded an F1-score of 0.78 for the ore class using the RFC algorithm. In the next step, the performance of the developed discriminative models was tested using hyperspectral data of two muck piles scanned in the open-pit gold mine. The results demonstrated the robustness of the mineral mapping method under field conditions compared to the FE method. These results highlight hyperspectral imaging as a valuable tool for improving ore-sorting efficiency in mining operations.

Enhancement of polychrome pottery hyperspectral images based on multilevel feature extraction method

Enhancement of polychrome pottery hyperspectral images based on multilevel feature extraction method

Performance of Radiomics and Deep Learning Models in Predicting Distant Metastases in Soft Tissue Sarcomas: A Systematic Review and Meta-analysis.

Performance of Radiomics and Deep Learning Models in Predicting Distant Metastases in Soft Tissue Sarcomas: A Systematic Review and Meta-analysis.

Mobile malware detection method using improved GhostNetV2 with image enhancement technique

In recent years, image-based feature extraction and deep learning classification methods are widely used in the field of malware detection, which helps improve the efficiency of automatic malicious feature extraction and enhances the overall performance of detection models. However, recent studies reveal that adversarial sample generation techniques pose significant challenges to malware detection models, as their effectiveness significantly declines when identifying adversarial samples. To address this problem, we propose a malware detection method based on an improved GhostNetV2 model, which simultaneously enhances detection performance for both normal malware and adversarial samples. First, Android classes.dex files are converted into RGB images, and image enhancement is performed using the Local Histogram Equalization technique. Subsequently, the Gabor method is employed to transform three-channel images into single-channel images, ensuring consistent detection accuracy for malicious code while reducing training and inference time. Second, we make three improvements to GhostNetV2 to more effectively identify malicious code, including introducing channel shuffling in the Ghost module, replacing the squeeze and excitation mechanism with a more efficient channel attention mechanism, and optimizing the activation function. Finally, extensive experiments are conducted to evaluate the proposed method. Results demonstrate that our model achieves superior performance compared to 20 state-of-the-art deep learning models, attaining detection accuracies of 97.7% for normal malware and 92.0% for adversarial samples.

Flatness pattern recognition based on stacked sparse denoising autoencoder and improved Osprey optimisation algorithm kernel-extreme learning machine

Aiming at the problems of random noise and insufficient extraction of flatness features in the data detected by the plate shaper during the rolling process of cold-rolled strip steel, this paper proposes a flatness recognition method based on stack sparse denoising autoencoder (SSDAE) with improved Osprey optimisation algorithm kernel-extreme learning machine (IOOA-KELM). The method first uses SSDAE to denoise and downscale the flatness data to achieve efficient feature extraction. Then, the regularisation coefficients and kernel parameters of KELM are optimised using IOOA to construct an accurate and efficient flatness recognition model. Verified by MATLAB simulation, the established model performs well in cold-rolled strip flatness recognition, with the mean value of root mean square error (RMSE) reduced to 0.0011, which improves the recognition performance by about 95% compared with that of traditional feature extraction methods. The experimental results show that the method has strong anti-interference ability, fast convergence speed and high recognition accuracy, which is a practical and efficient flatness pattern recognition method suitable for the quality needs in modern steel production.

Same data, different results? Machine learning approaches in bioacoustics

Abstract Automated acoustic analysis is increasingly used in behavioural ecology, and determining caller identity is a key element for many investigations. However, variability in feature extraction and classification methods limits the comparability of results across species and studies, constraining conclusions we can draw about the ecology and evolution of the groups under study. We investigated the impact of using different feature extraction (spectro‐temporal measurements, linear and Mel‐frequency cepstral coefficients (MFCC), as well as highly comparative time‐series analysis) and classification methods (discriminant function analysis, neural networks, random forests (RF), and support vector machines) on the consistency of caller identity classification accuracy across 16 mammalian datasets. We found that MFCCs and RFs yield consistently reliable results across datasets, facilitating a standardised approach across species that generates directly comparable data. These findings remained consistent across vocalisation sample sizes and number of individuals considered. We offer guidelines for processing and analysing mammalian vocalisations, fostering greater comparability and advancing our understanding of the evolutionary significance of acoustic communication in diverse mammalian species.

Exploring a Hybrid Convolutional Framework for Camouflage Target Classification in Land‐Based Hyperspectral Images

Exploring a Hybrid Convolutional Framework for Camouflage Target Classification in Land‐Based Hyperspectral Images

Leveraging data analytics for detection and impact evaluation of fake news and deepfakes in social networks

The past decade has seen a rapid and vast adoption of social media globally and over sixty percent of people were connected online through various social media platforms as of the start of 2024. Despite many advantages social media offers, one of the most significant challenges is the rapid rise of fake news and AI-generated deepfakes across these social networks. The spread of fake news and deepfakes can lead to a series of negative impacts, such as social trust, economic consequences, public health and safety crises, as demonstrated during the COVID-19 pandemic. Hence, it is more important now than ever to develop solutions to identify such fake news and deepfakes, and curb their spread. This paper begins with a review of the literature on the definitions of fake news and deepfakes, their different types and major differences, and the ways they spread. Building on this literature research, this paper aims to analyse how fake news can be identified using machine learning models, and understand how data analytics can be leveraged to evaluate the impact of such fake news on public behaviour and trust. A fake news detection framework is developed, where TF-IDF vectorization and bag of n-grams methods are implemented to extract text features, and six typical machine learning models are used to detect fake news, with the XGBoost classifier achieving the highest accuracy using both feature extraction methods. Additionally, a convolutional neural network model is designed to detect deepfake images with two distinct architectures, namely, ResNet50 and DenseNet121. To analyse the emotional impact of fake news on public behaviour and trust, a trained natural language toolkit called VADER lexicon is used to assign sentiment polarity and emotion strength to articles. The rampant rise of deepfake technology poses huge risks to social trust and privacy issues, which impacts both individuals and society at large, and leveraging the effective use of data analytics, machine learning and AI techniques can help prevent irreparable damage and mitigate the negative impacts of deepfakes in social networks. Finally, the paper discusses some practical solutions to mitigate the negative impacts of fake news and deepfakes.

Advancing Cancer Care: PET/CT Radiomics for Assessing Treatment Response to Chemoradiotherapy in Head and Neck Cancers

Head and neck cancers (HNC), comprising malignancies of the oral cavity, pharynx, and larynx, are among the most aggressive and functionally debilitating cancers worldwide. With over 8.5 lakh new cases annually, they present significant morbidity and mortality challenges. Concurrent chemoradiotherapy (CRT) remains the cornerstone of treatment for locally advanced HNC. However, timely and accurate assessment of treatment response is critical to guide subsequent management strategies such as treatment de-escalation, salvage surgery, or immunotherapy. Current assessment modalities, including Response Evaluation Criteria in Solid Tumors (RECIST) and qualitative [^18F]FDG-PET/CT interpretations, often suffer from interobserver variability, delayed changes post-treatment, and insufficient sensitivity in detecting subtle or early biological changes. This systematic review explores the role of radiomics-based analysis of PET/CT imaging in assessing treatment response to CRT in HNC patients. It aims to summarize current evidence on the prognostic and predictive value of radiomic features extracted from PET/CT, comparing them with conventional metrics such as SUVmax and RECIST-based evaluations. A comprehensive literature search was conducted using PubMed, Embase, and Scopus databases for studies published between January 2010 and March 2025. Inclusion criteria encompassed original research articles evaluating radiomic features extracted from pre-, mid-, or post-treatment PET/CT scans in HNC patients undergoing CRT. Studies reporting treatment response, disease-free survival (DFS), progression-free survival (PFS), or overall survival (OS) as endpoints were included. Twenty-eight eligible studies were included in the final analysis. Texture-based radiomic features derived from gray-level co-occurrence matrix (GLCM), gray-level run length matrix (GLRLM), and shape descriptors emerged as strong predictors of CRT response. Radiomic models demonstrated superior prognostic accuracy compared to SUVmax alone. However, heterogeneity in image acquisition protocols, feature extraction methods, and lack of external validation limited clinical applicability. Radiomics applied to PET/CT imaging holds promise as a non-invasive tool for individualized response assessment in HNC. Future multicentric prospective trials with standardized protocols and harmonized radiomics workflows are imperative for successful clinical translation.

Deep learning-based allergic rhinitis diagnosis using nasal endoscopy images

Allergic rhinitis typically has edematous and pale turbinates or erythematous and inflamed turbinates. While traditional approaches include using skin prick tests (SPT) to determine the presence of AR, It is often not related to actual symptoms, and it is an invasive test. We use deep learning to analyze nasal endoscopy images to investigate a quantitative method for diagnosing allergic rhinitis. Traditional machine learning-based diagnostic techniques have relied on structured clinical datasets featuring statistical data such as demographic characteristics, symptom severity, and clinical test results. In contrast, we propose a novel approach to use endoscopy image data to analyze the color distribution in the inferior turbinate region of patients with allergic rhinitis using the CIE-Lab color space and extract the adaptive histogram features that are used to explore and find suitable feature extraction methods and deep learning model architectures. Our proposed model achieves a promising diagnostic accuracy of 90.80% for images exhibiting AR symptoms. Future research will expand the dataset to include a broader spectrum of symptomatic and asymptomatic images to enhance model robustness and investigate the potential of optical analysis as a non-invasive diagnostic method for AR. This study introduced a novel approach to diagnosing allergic rhinitis using nasal endoscopy images. Our approach analyzed the color distribution of the inferior turbinates within the LAB color space, extracted important features from endoscopy images using both CNN feature extraction and histograms, and performed classification through SVM and fully connected classifiers.

A Feature Extraction Method for Parameter Fault Diagnosis of DC-DC Converters Under Strong Noise and Small Sample Conditions

Existing feature extraction methods struggle with low accuracy in strong noise and small sample scenarios, affecting parametric fault diagnosis in DC-DC converters. To address this issue, we propose an Adaptive Euler Difference Feature Extraction (AEDFE) method that extracts spatial features from fault signals to enhance the differentiation between parametric fault features of varying severity. This approach is implemented through a simple convolutional neural network, achieving high-precision diagnosis of DC-DC converter parametric faults even in challenging conditions. Experimental results demonstrate that the proposed AEDFE achieves 100% diagnostic accuracy in strong noise environments, with an average improvement of 61.61% compared to three other methods. Additionally, when training data is reduced to 10% of the original, the method still maintains an accuracy of 99.98%, representing a 77.64% increase in diagnostic precision compared to the comparative methods. These findings effectively demonstrate the superior performance of AEDFE in strong noise and small sample environments.

Recent advancements in feature extraction and classification based bone cancer detection – a systematic review

Cancer is a deadly disease that occurs due to the overgrowth of abnormal cells. Bone cancer is the third most occurring disease; approximately 10,000 patients suffer from bone cancer in India annually. It can lead to death if not diagnosed in the earlier stage. The bone cancer occurs in four stages as follows: In stage 1, cancer does not spread to other bone parts; in stage 2, cancer looks similar to stage 1, but it becomes dangerous; in stage 3, cancer spreads to one or two bone parts; and in stage 4, cancer spreads to other body parts. Timely diagnosis of bone cancer is challenging due to the unspecific indications that are similar to common musculoskeletal injuries, late visits of patients to the hospital, and low intuition by the physician. The texture of diseased bone differs from healthy bone. Mostly in the dataset, the healthy and cancerous bone images have similar characteristics. Therefore, development of automated systems is necessary to classify the normal and abnormal scan images. The objective of this paper is to identify the studies on classification techniques in detecting bone cancer with five criteria: feature extraction methods, machine learning (ML) and deep learning (DL) techniques, advantages, disadvantages, and classifier accuracy. The current study performed the systematic literature review of 129 studies selected based on the use of different feature extractions to extract the textural characteristics of the images that are fed into the ML and DL algorithms to classify the normal and subtypes of bone cancer images for better analysis. The review concludes that the convolutional neural network classifier, along with different textural feature extraction techniques like gray-level co-occurrence matrix (GLCM) and local binary pattern (LBP), detected the bone cancer with high accuracy compared to DL classification without feature extraction techniques in diagnosing the bone cancer. In this respect, this paper proposes the systematic review of types of bone cancer and recent advancements in feature extraction methods and classification involving deep learning and machine learning models to detect bone cancer with a higher accuracy rate.

EEG Signal Analysis for the Classification of Alzheimer's and Frontotemporal Dementia: A Novel Approach Using Artificial Neural Networks and Cross-Entropy Techniques

Dementia, a neurological disorder, can cause cognitive decline due to damage to the brain. Our study aims to contribute to the development of computer-aided diagnosis (CAD) systems to aid in the early diagnosis of Alzheimer's disease (AD) and frontotemporal dementia (FTD) by examining Electroencephalogram (EEG) signals. EEG signals of 36 AD, 23 FTD and 29 healthy control (HC) participants were analysed and entropy measurement approaches were used to analyze the connectivity between EEG channel pairs. The Cross Permutation Entropy (CPE) method and the Cross Conditional Entropy (CCE) method were analysed separately and the Fused Cross Entropy (FCE) method was also tested by combining these techniques to determine the most appropriate method for feature extraction from EEG signals. The features obtained from these techniques were then evaluated in the classification phase using two separate machine learning algorithms. According to the performance evaluation criteria, the FCE and Artificial Neural Network (ANN) model showed the most successful performance in the classification of all groups. In terms of Area Under the Curve (AUC) and accuracy rates, 99.85% AUC and 98.46% accuracy were obtained in AD&HC groups, 99.71% AUC and 98.10% accuracy in FTD&HC groups, and 99.39% AUC, 96.61% accuracy in AD&FTD groups. With the same model, an AUC rate of 97.14% and accuracy rate of 73.86% was obtained for the classification of the triple group (AD&FTD&HC). It has been observed that the results of this study show successful performance compared to the results of similar studies.

Feature Extraction and Diagnosis of Power-Shift System Faults in Unmanned Hydro-Mechanical Transmission Tractors

To enhance the reliability of unmanned hydro-mechanical transmission tractors, a fault diagnosis method for their power-shift system was developed. First, fault types were identified, and sample data was collected via a test bench. Next, a feature extraction method for data dimensionality reduction and a deep learning network called W_SCBAM were introduced for fault diagnosis. Both W_SCBAM and conventional algorithms were trained 20 times, and their performance was compared. Further testing of W_SCBAM was conducted in various application scenarios. The results indicate that the feature extraction method reduces the sample length from 46 to 3. The fault diagnosis accuracy of W_SCBAM for the radial-inlet clutch system has an expectation of 98.5% and a variance of 1.6%, respectively, outperforming other algorithms. W_SCBAM also excels in diagnosing faults in the axial-inlet clutch system, achieving 97.6% accuracy even with environmental noise. Unlike traditional methods, this study integrates the update of a dimensionality reduction matrix into network parameter training, achieving high-precision classification with minimal input data and lightweight network structure, ensuring reliable data transmission and real-time fault diagnosis of unmanned tractors.

A novel Probabilistic Bi-Level Teaching–Learning-Based Optimization (P-BTLBO) algorithm for hybrid feature extraction and multi-class brain tumor classification using ResNet-50 and GLCM

The efficient and precise classification of brain tumors is important for early identification and appropriate action. This study presents a novel technique called Probabilistic Bi-Level Teaching–Learning-Based Optimization (P-BTLBO) for hybrid feature extraction and multi-class brain tumor classification. The P-BTLBO method combines probabilistic modeling with a bi-level optimization framework to make feature selection better. This makes it easier to explore and exploit in environments with a lot of dimensions. Deep features from ResNet-50, a convolutional neural network, are combined with texture-based features from gray level co-occurrence matrix (GLCM) analysis in the hybrid feature extraction method. These complementary attributes capture both predominant patterns and detailed texture information from magnetic resonance imaging (MRI) scans, facilitating thorough tumor characterization. The suggested P-BTLBO method makes the combined set of features better by repeatedly focusing on higher-level and lower-level goals: improving classification accuracy while removing unnecessary features. To assess the efficacy of the optimized features, various classifiers, such as support vector machine (SVM), k-nearest neighbors (k-NN), and decision tree, were examined. Experimental findings indicate that the P-BTLBO algorithm surpasses conventional optimization methods, including TLBO and PSO, regarding classification accuracy, feature subset size, and computational efficiency. The hybrid framework attains enhanced multi-class categorization of glioma, meningioma, pituitary tumors, and healthy cases, presenting a valuable instrument for clinical diagnosis. This research underscores the efficacy of P-BTLBO in tackling hierarchical optimization issues in medical imaging and outlines its applicability in other intricate fields.

A Novel Deep Learning Model for Motor Imagery Classification in Brain–Computer Interfaces

Recent advancements in decoding electroencephalogram (EEG) signals for motor imagery tasks have shown significant potential. However, the intricate time–frequency dynamics and inter-channel redundancy of EEG signals remain key challenges, often limiting the effectiveness of single-scale feature extraction methods. To address this issue, we propose the Dual-Branch Blocked-Integration Self-Attention Network (DB-BISAN), a novel deep learning framework for EEG motor imagery classification. The proposed method includes a Dual-Branch Feature Extraction Module designed to capture both temporal features and spatial patterns across different scales. Additionally, a novel Blocked-Integration Self-Attention Mechanism is employed to selectively highlight important features while minimizing the impact of redundant information. The experimental results show that DB-BISAN achieves state-of-the-art performance. Also, ablation studies confirm that the Dual-Branch Feature Extraction and Blocked-Integration Self-Attention Mechanism are critical to the model’s performance. Our approach offers an effective solution for motor imagery decoding, with significant potential for the development of efficient and accurate brain–computer interfaces.

Sequence-similarity-based approach to SARS–CoV-2 genome sequence and lung cancer-related genes via multivariate feature extraction method

The COVID-19 pandemic has prompted genomic studies linking SARS–CoV-2 and lung cancer-related genes. This study explores sequence similarity and motif patterns to assess disease susceptibility. We applied a data mining approach to compare human and SARS–CoV-2 genomes, revealing high sequence identity (0.74–0.99%) with lung cancer-related genes. Low-entropy motifs were associated with higher genetic risk. We identified shared patterns of lengths 4, 5, and 10, selecting the most significant motifs. These findings support the hypothesis that sequence similarity and conserved motifs provide insights into gene function, evolutionary processes, and the genetic links between cancer and viral infections.

Sentiment Analysis of Product Reviews Using LSTM - A Comparative Evaluation with Machine Learning Algorithms Employing BOW and TF-IDF Techniques

Sentiment analysis has become an invaluable tool in understanding consumer opinions in large datasets. This study explores sentiment analysis of the product review dataset applying different machine learning classification algorithms, specifically focusing on two primary feature extraction methods: (TF-IDF) and (BOW) A thorough comparison was conducted to assess the effectiveness of each method alone, as well as a novel hybrid technique that merges both TF-IDF and BOW. And compared with deep learning approach, our findings demonstrate that feature extraction technique significantly enhances classification performance. Among the tested algorithms, logistic regression with tfidf, bow exhibited even greater accuracy. Obtaining the most accurate results possible from the sentiment analysis is the primary objective of this endeavor. The first step in the process of analyzing and classifying the data is going to be the preprocessing of the data, followed by the extraction of features, then the categorization of sentiments via the use of machine learning algorithms, and lastly the assessment of the algorithms. The end findings indicate that the SVM classifier obtained an accuracy of 93%, the Naive Bayes classifier achieved an accuracy of 91%, the Logistic regression classifier got an accuracy of 94%, and the LSTM classifier earned an accuracy which was 93.58%. In future work may explore the integration of additional feature extraction methods with deep learning to refine and improve sentiment analysis models.

Multi‐Objective Manifold Representation for Opinion Mining

ABSTRACTSentiment analysis plays a crucial role across various domains, requiring advanced methods for effective dimensionality reduction and feature extraction. This study introduces a novel framework, multi‐objective manifold representation (MOMR) for opinion mining, which uniquely integrates deep global features with local manifold representations to capture comprehensive data patterns efficiently. Unlike existing methods, MOMR employs advanced dimensionality reduction techniques combined with a self‐attention mechanism, enabling the model to focus on contextually relevant textual elements. This dual approach not only enhances interpretability but also improves the performance of sentiment analysis. The proposed method was rigorously evaluated against both classical techniques such as long short‐term memory (LSTM), naive Bayes (NB) and support vector machines (SVMs), and modern state‐of‐the‐art models including recurrent neural networks (RNN) and convolutional neural networks (CNN). Experiments on diverse datasets: IMDB, Fake News, Twitter and Yelp demonstrated the superior accuracy and robustness of MOMR. By outperforming competing methods in terms of generalizability and effectiveness, MOMR establishes itself as a significant advancement in sentiment analysis, with broad applicability in real‐world opinion mining tasks (https://github.com/pshtirahman/Sentiment‐Analysis.git).

An Enhanced Automatic Lung Disease Diagnosis Scheme Using ECG Signals with Integrated Feature Extraction and Improved Deep Learning

An early detection of lung disease can avoid patient death by giving useful treatment. The human with related lung conditions nearly contains related electrocardiogram (ECG) signals. The ECG examination can be an analytical system employed on the screen for various lung diseases. Arrhythmias are discovered through patterns of ECG signals. Nowadays, most of the ECG analysis is done according to the medical team‟s personal opinion, which may have led to more burden. Therefore, in this paper, an automatic lung disease diagnosis scheme is presented through an accurate ECG signal categorization using improved deep learning processes. Initially, ECG signal data is pre-processed with noise removal and QRS complex discovery schemes. Subsequently, an integrated feature extraction method is proposed in this paper to extract the ECG wave features. The presented automatic lung disease detection scheme is examined using the ECG signals dataset collected from a MIT-BIH arrhythmia database.