Automatic Feature Extraction Research Articles

Smart Distillation for Land Use Monitoring: A Lightweight Remote Sensing Segmentation Strategy

As the demand for land use monitoring continues to grow, high-precision remote sensing products have become increasingly important. Compared to traditional methods, deep learning networks demonstrate significant advantages in automatic feature extraction, handling complex scenes, and improving classification accuracy. However, as the complexity of these networks increases, so does the computational cost. To address this challenge, we propose an innovative knowledge distillation model, integrating two key modules—spatial-global attention feature distillation (SGAFD) and channel attention-based relational distillation (CARD). This model enables a lightweight “student” network to be guided by a large “teacher” network, enhancing classification performance while maintaining a compact model size. We validated our approach on the large-scale public remote sensing datasets GID15 and LoveDA, and the results show that these modules effectively improve classification performance, overcoming the limitations of lightweight models and advancing the practical applications of land use monitoring.

Ultra-Short-Term Photovoltaic Power Prediction Based on BiLSTM with Wavelet Decomposition and Dual Attention Mechanism

Photovoltaic power generation relies on sunlight conditions, and traditional prediction models find it difficult to capture the deep features of power data, resulting in low prediction accuracy. In addition, there are problems such as outliers and missing values in the data collected on site. This article proposes an ultra-short-term photovoltaic power generation prediction model based on wavelet decomposition, a dual attention mechanism, and a bidirectional long short-term memory network (W-DA-BiLSTM), aiming to address the limitations of existing deep learning models in processing nonlinear data and automatic feature extraction and optimize for the common problems of outliers and missing values in on-site data collection. This model uses the quartile range method for outlier detection and multiple interpolation methods for missing value completion. In the prediction section, wavelet decomposition is used to effectively handle the volatility and nonlinear characteristics of photovoltaic power generation data, while the bidirectional long short-term memory network (LSTM) structure and dual attention mechanism enhance the model’s comprehensive learning ability for time series data. The experimental results show that compared with the SOTA method, the model proposed in this paper has higher accuracy and efficiency in predicting photovoltaic power generation and can effectively address common random fluctuations and nonlinear problems in photovoltaic power generation.

MEPFeatX—automated feature extraction of motor-evoked potentials in transcranial magnetic stimulation

Motor evoked potentials (MEPs) are an important measure in transcranial magnetic stimulation (TMS) when assessing neuronal excitability in clinical diagnostics related to motor function, as well as in neuroscience research. However, manual feature extraction from large datasets can be time-consuming and prone to human error, and valuable features, such as MEP polyphasia and duration, are often neglected. Several packages have been developed to simplify the process; however, they are often tailored to specific studies or are not accessible. Here, we introduce MEPFeatX, a verified MATLAB package designed for automated and comprehensive MEP feature extraction across a wide range of stimulation paradigms. MEPFeatX is designed and documented for easy integration into any MEP analysis pipeline. Primed templates for specific paradigms, as well as additional analysis coded in R language, are also provided. Thus, MEPFeatX provides its users with a comprehensive and accurate set of MEP features, along with their visuals, facilitating quick and reliable MEP analysis in TMS studies.

SC-ResNeXt: A Regression Prediction Model for Nitrogen Content in Sugarcane Leaves

In agricultural production, the nitrogen content of sugarcane is assessed with precision and the economy, which is crucial for balancing fertilizer application, reducing resource waste, and minimizing environmental pollution. As an important economic crop, the productivity of sugarcane is significantly influenced by various environmental factors, especially nitrogen supply. Traditional methods based on manually extracted image features are not only costly but are also limited in accuracy and generalization ability. To address these issues, a novel regression prediction model for estimating the nitrogen content of sugarcane, named SC-ResNeXt (Enhanced with Self-Attention, Spatial Attention, and Channel Attention for ResNeXt), has been proposed in this study. The Self-Attention (SA) mechanism and Convolutional Block Attention Module (CBAM) have been incorporated into the ResNeXt101 model to enhance the model’s focus on key image features and its information extraction capability. It was demonstrated that the SC-ResNeXt model achieved a test R2 value of 93.49% in predicting the nitrogen content of sugarcane leaves. After introducing the SA and CBAM attention mechanisms, the prediction accuracy of the model improved by 4.02%. Compared with four classical deep learning algorithms, SC-ResNeXt exhibited superior regression prediction performance. This study utilized images captured by smartphones combined with automatic feature extraction and deep learning technologies, achieving precise and economical predictions of the nitrogen content in sugarcane compared to traditional laboratory chemical analysis methods. This approach offers an affordable technical solution for small farmers to optimize nitrogen management for sugarcane plants, potentially leading to yield improvements. Additionally, it supports the development of more intelligent farming practices by providing precise nitrogen content predictions.

An Improved Method for Human Activity Detection with High-Resolution Images by Fusing Pooling Enhancement and Multi-Task Learning

Deep learning has garnered increasing attention in human activity detection due to its advantages, such as not relying on expert knowledge and automatic feature extraction. However, the existing deep learning-based approaches are primarily confined to recognizing specific types of human activities, which hinders scientific decision-making and comprehensive environmental protection. Therefore, there is an urgent need to develop a deep learning model to address multiple-type human activity detection with finer-resolution images. In this study, we proposed a new multi-task learning model (named PE-MLNet) to simultaneously achieve change detection and land use classification in GF-6 bitemporal images. Meanwhile, we also designed a pooling enhancement module (PEM) to accurately capture multi-scale change details from the bitemporal feature maps through combining differencing and concatenating branches. An independent annotated dataset at Yellow River Delta was taken to examine the effectiveness of PE-MLNet. The results showed that PE-MLNet exhibited obvious improvements in both detection accuracy and detail handling compared with other existing methods. Further analysis uncovered that the areas of buildings, roads, and oil depots has obviously increased, while the farmland and wetland areas largely decreased over the five years, indicating an expansion of human activities and their increased impacts on natural environments.

Advanced algorithm for step detection in single-entity electrochemistry: a comparative study of wavelet transforms and convolutional neural networks.

Single-entity electrochemistry (SEE) is an emerging field within electrochemistry focused on investigating individual entities such as nanoparticles, bacteria, cells, or single molecules. Accurate identification and analysis of SEE signals require effective data processing methods for unbiased and automated feature extraction. In this study, we apply and compare two approaches for step detection in SEE data: discrete wavelet transforms (DWT) and convolutional neural networks (CNN).

An Automatic Feature Extraction Method for Gas Sensors Based on Color-Enhanced Phase Space

An Automatic Feature Extraction Method for Gas Sensors Based on Color-Enhanced Phase Space

Archimedes assisted LSTM model for blockchain based privacy preserving IoT with smart cities

Presently, the emergence of internet of things (IoT) has significantly improved the processing, analysis, and management of the substantial volume of big data generated by smart cities. Among the various applications of smart cities, notable ones include location-based services, urban design and transportation management. These applications, however, come with several challenges, including privacy concerns, mining complexities, visualization issues and data security. The integration of blockchain (BC) technology into IoT (BIoT) introduces a novel approach to secure smart cities. This work presents an Archimedes assisted long short-term memory (LSTM) model intrusion detection for BC based privacy preserving (PP) IoT with smart cities. After the stage of pre-processing, the LSTM is utilized for automated feature extraction and classification. At last, the Archimedes optimizer (AO) is utilized to optimize the LSTM’s hyper-parameters. In addition, the BC technology is utilized for securing the data transmission.

Deep Learning Classification of Epileptic Magnetoencephalogram

In this research we study several statistical methods for feature extraction from Magnetoencephalography (MEG) Signals and classification of these signals to two classes: epileptic and healthy, based on the extracted features. We, then, apply automated feature extraction techniques by means of deep learning using several Artificial Neural Network (ANN) models. Our goal is to try various methods and models for MEG Signal classification and draw some conclusions about their functionality and effectiveness. We base our study on our theoretical knowledge on the neurology of epilepsy, previous studies of epileptic seizure imaging and recognition using MEG and EEG as well as the Signal Processing Theory and techniques. We apply several advanced classification methods with the use of ANN like Feed-Forward ANN, Convolutional NN and Inception V3. The results of this study are very encouraging and can be a base for future research on the subject of epileptic seizure recognition, prediction and prevention.

Facial expression recognition with modified local ternary pattern images using convolutional neural network and extreme learning machine

Facial emotion recognition is a crucial aspect of human-computer interaction systems, education, health and various other fields. The common challenges faced in FER are variation in light. Based on researches, Local ternary Pattern (LTP) as a feature vector that can handle the variation of light. Modified Local Ternary Pattern (MLTP) is more discriminative to variation of light and less sensitive to noise. MLTP is a conventional feature vector that requires manual processing. In contrast to MLTP, the Convolution Neural Network (CNN) architecture has an automated feature extractor. This paper proposes MLTP as input to the CNN architecture for handling the variation of light and for automatic feature extractor. Then, for classifying Extreme Learning Machine is used to reduce the training time of the CNN classifier. The Karolinska Directed Emotional Faces (KDEF) dataset is used to assess the proposed method and got an accuracy of 86.28%.

Machine Learning in Brain Tumor Diagnosis: Assessing the Efficacy of Diverse Methods

Background: Brain tumors constitute a critical and potentially life-threatening condition that requires accurate and timely diagnosis. In recent years, machine learning (ML) approaches have emerged as powerful tools for assisting radiologists and clinicians in identifying and classifying tumors on medical imaging. While numerous ML methods, including conventional machine learning algorithms and deep learning-based models, have been explored, a comprehensive analysis of their efficacy in detecting, segmenting, and classifying brain tumors remains essential. Methods: We conducted a systematic evaluation of diverse ML methods used in brain tumor diagnosis. We first identified and collated relevant studies from major scientific databases, focusing on image-based applications such as magnetic resonance imaging (MRI). The methods included pre-processing pipelines, feature extraction techniques, and both supervised and unsupervised learning approaches. We then compared these methods based on standard metrics, including accuracy, sensitivity, specificity, and F1-score, to gain insights into their relative performance. Furthermore, we applied representative algorithms (such as support vector machines, convolutional neural networks, and random forests) to an open-access brain tumor imaging dataset to evaluate their performance and compare empirical findings. Results: Results indicated that deep learning models, particularly convolutional neural networks, consistently outperformed traditional ML models across various performance metrics. In particular, automated feature extraction in deep learning approaches proved pivotal in capturing nuanced information such as tumor shape and heterogeneity. Conventional algorithms still demonstrated merit when dataset sizes were limited or when computational resources were constrained. Additionally, various data augmentation techniques improved the robustness and generalizability of ML models in situations with scarce annotated data. Conclusion: Our findings suggest that deep learning-based methods hold the greatest potential for accurate and efficient brain tumor diagnosis, particularly when coupled with robust datasets and high-quality imaging. Nonetheless, careful selection of model architecture, training strategies, and validation protocols remains paramount to ensure reliable clinical translation.

MS-CLSTM: Myoelectric Manipulator Gesture Recognition Based on Multi-Scale Feature Fusion CNN-LSTM Network.

Surface electromyography (sEMG) signals reflect the local electrical activity of muscle fibers and the synergistic action of the overall muscle group, making them useful for gesture control of myoelectric manipulators. In recent years, deep learning methods have increasingly been applied to sEMG gesture recognition due to their powerful automatic feature extraction capabilities. sEMG signals contain rich local details and global patterns, but single-scale convolutional networks are limited in their ability to capture both comprehensively, which restricts model performance. This paper proposes a deep learning model based on multi-scale feature fusion-MS-CLSTM (MS Block-ResCBAM-Bi-LSTM). The MS Block extracts local details, global patterns, and inter-channel correlations in sEMG signals using convolutional kernels of different scales. The ResCBAM, which integrates CBAM and Simple-ResNet, enhances attention to key gesture information while alleviating overfitting issues common in small-sample datasets. Experimental results demonstrate that the MS-CLSTM model achieves recognition accuracies of 86.66% and 83.27% on the Ninapro DB2 and DB4 datasets, respectively, and the accuracy can reach 89% in real-time myoelectric manipulator gesture prediction experiments. The proposed model exhibits superior performance in sEMG gesture recognition tasks, offering an effective solution for applications in prosthetic hand control, robotic control, and other human-computer interaction fields.

A Review of Fall Detection Research Based on Deep Learning

In recent years, with the intensification of societal aging, falls have emerged as one of the primary health risks threatening the well-being of the elderly, making the effective detection of fall events an urgent problem to address. Fall detection technology based on deep learning has increasingly become a focal point of research due to its powerful automatic feature extraction capabilities. This paper systematically reviews the recent research in fall detection, categorizing the technologies into two main types according to the devices used for data collection: image-based and non-image-based detection methods. The study highlights recent advances in fall detection using Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Temporal Convolutional Networks (T-CNN) frameworks, examining the application and advantages of these methods in processing various types of data, such as video sequences and sensor data. Through an in-depth analysis of the operational mechanisms of multiple deep learning architectures and their performance on different datasets, this paper elucidates the latest advancements and challenges facing current fall detection technologies. Additionally, commonly used fall detection datasets are introduced, including their data scales and application contexts. Finally, this paper provides a summary of recent progress in fall detection techniques, outlines future directions, and presents recommendations for advancement across several dimensions.

Offshore Wind‐Hydrogen Systems Fault Detection Based on CNN‐BiLSTM‐AM Algorithm

ABSTRACTThis study presents a novel deep learning‐based approach for fault detection in offshore wind‐hydrogen systems. A fault detection model is developed using convolutional neural networks (CNNs), bidirectional long short‐term memory networks (BiLSTMs), and an attention mechanism (AM). The model is trained on a dataset generated through fault injection techniques, which simulate real‐world faults in the system. Key operational parameters, such as wind speed and hydrogen production rate, are used to detect faults effectively. This paper reduces reliance on actual experiments, and introducing artificial faults allows system performance assessment under different fault scenarios, lowering project risks and costs. This work facilitates automatic feature extraction and high‐precision classification of time‐series fault data, which covers a fully automated learning process from data to fault detection. The outstanding performance of the method is validated through computation and result comparison.

An intelligent surface roughness prediction method based on automatic feature extraction and adaptive data fusion

Machining quality prediction based on cutting big data is the core focus of current developments in intelligent manufacturing. Presently, predictions of machining quality primarily rely on process and signal analyses. Process-based predictions are generally constrained to the development of rudimentary regression models. Signal-based predictions often require large amounts of data, multiple processing steps (such as noise reduction, principal component analysis, modulation, etc.), and have low prediction efficiency. In addition, the accuracy of the model depends on tedious manual parameter tuning. This paper proposes a convolutional neural network quality intelligent prediction model based on automatic feature extraction and adaptive data fusion (CNN-AFEADF). Firstly, by processing signals from multiple directions, time-frequency domain images with rich features can be obtained, which significantly benefit neural network learning. Secondly, the corresponding images in three directions are fused into one image by setting different fusion weight parameters. The optimal fusion weight parameters and window length are determined by the Particle Swarm Optimization algorithm (PSO). This data fusion method reduces training time by 16.74 times. Finally, the proposed method is verified by various experiments. This method can automatically identify sensitive data features through neural network fitting experiments and optimization, thereby eliminating the need for expert experience in determining the significance of data features. Based on this approach, the model achieves an average relative error of 2.95%, reducing the prediction error compared to traditional models. Furthermore, this method enhances the intelligent machining level.

Development of an Integrated AI Model Based on CNN-SVM for Electricity Theft Detection

This research presents the development and implementation of an integrated artificial intelligence model for electricity theft detection, combining Convolutional Neural Networks (CNN) and Support Vector Machines (SVM). The primary objective was to create a more accurate, efficient, and scalable method for identifying fraudulent electricity consumption patterns. Our CNN-SVM hybrid model leverages CNNs for automatic feature extraction from complex consumption data and SVMs for effective classification. This synergy allows for superior performance in detecting subtle anomalies indicative of electricity theft. The methodology involved pre-processing a large dataset of electricity consumption records, training the CNN to extract relevant features, and optimising the SVM classifier to distinguish between normal and fraudulent patterns. We evaluated the model's performance using metrics including accuracy, precision, recall, F1-score, and ROC AUC. Results demonstrated that our integrated CNN-SVM model significantly outperformed conventional machine learning techniques and standalone models in electricity theft detection. The model achieved an accuracy of 96.6%, with a precision of 97.2% and a recall of 96.1%. Comparative analysis against other state-of-the-art algorithms revealed consistently superior performance across all evaluation metrics. To enhance practical applicability, we developed and deployed a web application that implements the model, allowing for user-friendly interaction and real-time theft detection. This addition bridges the gap between research and real-world implementation, providing utility companies with an accessible tool for fraud detection. The study also explored the model's potential for real-time application and scalability to large-scale utility operations. Our findings suggest that the computational efficiency of the CNN-SVM model, coupled with the web application, offers utility companies a powerful and accessible tool for rapid response to potential fraud. This research contributes to the field of electricity theft detection by introducing a novel, high-performance AI model with a practical web-based implementation. The proposed approach not only improves detection accuracy but also offers potential for immediate real-world application, paving the way for more effective fraud prevention in the utility sector.

An Efficient Deep Learning Approach for Malaria Parasite Detection in Microscopic Images.

Background: Malaria is a life-threatening disease spread by infected mosquitoes, affecting both humans and animals. Its symptoms range from mild to severe, including fever, muscle discomfort, coma, and kidney failure. Accurate diagnosis is crucial but challenging, relying on expert technicians to examine blood smears under a microscope. Conventional methods are inefficient, while machine learning approaches struggle with complex tasks and require extensive feature engineering. Deep learning, however, excels in complex tasks and automatic feature extraction. Objective: This paper presents EDRI, which is a novel hybrid deep learning model that integrates multiple architectures for malaria detection from red blood cell images. The EDRI model is designed to capture diverse features and leverage multi-scale analysis. Methods: The proposed EDRI model is trained and evaluated on the NIH Malaria dataset comprising 27,558 labeled microscopic red blood cell images. Results: Experiments demonstrate its effectiveness, achieving an accuracy of 97.68% in detecting malaria, making it a valuable tool for clinicians and public health professionals. Conclusions: The results demonstrate the effectiveness of proposed model's ability to detect malaria parasite from red blood cell images, offering a robust tool for rapid and reliable malaria diagnosis.

Convolution spatial-temporal attention network for EEG emotion recognition

In recent years, emotion recognition using electroencephalogram (EEG) signals has garnered significant interest due to its non-invasive nature and high temporal resolution. We introduced a groundbreaking method that bypasses traditional manual feature engineering, emphasizing data preprocessing and leveraging the topological relationships between channels to transform EEG signals from two-dimensional time sequences into three-dimensional spatio-temporal representations. Maximizing the potential of deep learning, our approach provides a data-driven and robust method for identifying emotional states. Leveraging the synergy between convolutional neural network and attention mechanisms facilitated automatic feature extraction and dynamic learning of inter-channel dependencies. Our method showcased remarkable performance in emotion recognition tasks, confirming the effectiveness of our approach, achieving average accuracy of 98.62% for arousal and 98.47% for valence, surpassing previous state-of-the-art results of 95.76% and 95.15%. Furthermore, we conducted a series of pivotal experiments that broadened the scope of emotion recognition research, exploring further possibilities in the field of emotion recognition.

A Comparative Analysis of Convolutional Neural Networks for American Sign Language Recognition

Abstract. Sign language recognition is an important technology that makes it possible for ordinary people to be able to communicate with deaf people, fostering inclusivity and accessibility. The came out of deep learning technology has completely changed the field by enabling the automatic extraction and learning of hierarchical features from raw data, leading to significant improvements in recognition accuracy. This paper presents a comprehensive comparative analysis of different Convolutional Neural Network (CNN) architectures for recognizing American Sign Language (ASL) signs. Utilizing the sign language dataset, which contains 24 classes of ASL letters represented by 28x28 grayscale images, the author evaluated the performance of a Basic CNN, a Modified Residual Network (ResNet)-50, and a LeNet-5 model. This study emphasizes the impact of architectural choices on recognition accuracy and computational efficiency. Results indicate that while ResNet-50 demonstrates superior accuracy, fluctuating significantly during initial training, the Basic CNN and LeNet-5 models offer greater stability with slightly lower accuracy. This work concludes that despite the initial challenges, deep learning models, particularly ResNet-50, show promise for ASL recognition, highlighting the need for diverse and enriched datasets to improve model reliability in real-world scenarios.

GMN+: A Binary Homologous Vulnerability Detection Method Based on Graph Matching Neural Network with Enhanced Attention

The widespread reuse of code in the open-source community has led to the proliferation of homologous vulnerabilities, which are security flaws propagated across diverse software systems through the reuse of vulnerable code. Such vulnerabilities pose serious cybersecurity risks, as attackers can exploit the same weaknesses across multiple platforms. Deep learning has emerged as a promising approach for detecting homologous vulnerabilities in binary code due to their automated feature extraction and high efficiency. However, existing deep learning methods often struggle to capture deep semantic features in binary code, limiting their effectiveness. To address this limitation, this paper presents GMN+, which is a novel graph matching neural network with enhanced attention for detecting homologous vulnerabilities. This method comprehensively considers the information contained in instructions and incorporates types of input instruction. Masked Language Modeling and Instruction Type Prediction are developed as pre-training tasks to enhance the ability of GMN+ in extracting semantic information from basic blocks. GMN+ utilizes an attention mechanism to focus concurrently on the critical semantic information within functions and differences between them, generating robust function embeddings. Experimental results indicate that GMN+ outperforms state-of-the-art methods in various tasks and achieves notable performance in real-world vulnerability detection scenarios.