Convolutional Recurrent Neural Network Research Articles

Applying Machine Learning and Bayesian Inference to Identify and Locate Moving Anthropogenic Sources Using Distributed Acoustic Sensing Data

Abstract Distributed acoustic sensing (DAS) systems, which use existing telecommunication fibers, offer high-resolution capabilities ideal for recording anthropogenic sources. However, the complexity of urban environments and the large amount of data recorded by DAS require automated methods to efficiently detect and categorize anthropogenic sources. We evaluate how well three machine learning models (k-nearest neighbor [k-NN], convolutional neural networks, and recurrent-convolutional neural networks) can identify various anthropogenic sources recorded by DAS. Our findings reveal that both k-NN and neural network methods perform well in high signal-to-noise ratio (SNR) settings. However, their accuracy decreases at SNRs <4. We also use Kalman filtering, a form of Bayesian inference, on backprojected locations of these sources to recover locations that generally fall within standard smartphone Global Positioning System errors. By combining machine learning and Kalman filter results, we calculate a multidimensional model of moving anthropogenic sources. These results demonstrate the potential of DAS data in urban seismology for accurately identifying and locating such sources. Depending on the research objectives, these sources can be further studied or filtered out to improve the quality of seismic data for earthquake studies. Such methods provide a valuable tool for urban seismology and seismic hazard analysis.

Mifnet: a MamBa-based interactive frequency convolutional neural network for motor imagery decoding.

Motor imagery (MI) decoding remains a critical challenge in brain-computer interface (BCI) systems due to the low signal-to-noise ratio, non-stationarity, and complex spatiotemporal dynamics of electroencephalography (EEG) signals. Although deep learning architectures have advanced MI-EEG decoding, existing approaches-including convolutional neural networks (CNNs), Transformers, and recurrent neural networks (RNNs)-still face limitations in capturing global temporal dependencies, maintaining positional coherence, and ensuring computational efficiency. To address these challenges, we propose MIFNet, a MamBa-based Interactive Frequency Convolutional Neural Network that systematically integrates spectral, spatial, and temporal feature extraction. Specifically, MIFNet incorporates: non-overlapping frequency decomposition, which selectively extracts motor imagery-related mu (8-12 Hz) and beta (12-32 Hz) rhythms; a ConvEncoder module, which autonomously learns to fuse spectral-spatial features from both frequency bands; and a MamBa-based temporal module, leveraging selective state-space models (SSMs) to efficiently capture long-range dependencies with linear complexity. Extensive experiments on three public MI-EEG datasets (BCIC-IV-2A, OpenBMI, and High Gamma) demonstrate that MIFNet outperforms existing models, achieving an average classification accuracy improvement of 12.3%, 8.3%, 4.7%, and 5.5% over EEGNet, FBCNet, IFNet, and Conformer, respectively. Ablation studies further validate the necessity of each component, with the MamBa module contributing a 5.5% improvement in accuracy on the BCIC-IV-2A dataset. Moreover, MIFNet exhibits strong generalization performance in cross-validation settings, establishing a robust foundation for real-time BCI applications. Our findings highlight the potential of hybridizing CNNs with state-space models (SSMs) for improving EEG decoding performance, effectively bridging the gap between localized feature extraction and global temporal modeling.

Performance Analysis of a Hybrid Complex-Valued CNN-TCN Model for Automatic Modulation Recognition in Wireless Communication Systems

This paper presents a novel deep learning-based automatic modulation recognition (AMR) model, designed to classify ten modulation types from complex I/Q signal data. The proposed architecture, named CV-CNN-TCN, integrates Complex-Valued Convolutional Neural Networks (CV-CNNs) with Temporal Convolutional Networks (TCNs) to jointly extract spatial and temporal features while preserving the inherent phase information of the signal. An enhanced variant, CV-CNN-TCN-DCC, incorporates dilated causal convolutions to further strengthen temporal representation. The models are trained and evaluated on the benchmark RadioML2016.10b dataset. At SNR = −10 dB, the CV-CNN-TCN achieves a classification accuracy of 37%, while the CV-CNN-TCN-DCC improves to 40%. In comparison, ResNet reaches 33%, and other models such as CLDNN (convolutional LSTM dense neural network) and SCRNN (Sequential Convolutional Recurrent Neural Network) remain below 30%. At 0 dB SNR, the CV-CNN-TCN-DCC achieves a Jaccard index of 0.58 and an MCC of 0.67, outperforming ResNet (0.55, 0.64) and CNN (0.53, 0.61). Furthermore, the CV-CNN-TCN-DCC achieves 75% accuracy at SNR = 10 dB and maintains over 90% classification accuracy for SNRs above 2 dB. These results demonstrate that the proposed architectures, particularly with dilated causal convolutional enhancements, significantly improve robustness and generalization under low-SNR conditions, outperforming state-of-the-art models in both accuracy and reliability.

The singing style of female roles in ethnic opera under artificial intelligence and deep neural networks

With the rapid advancement of artificial intelligence technology, efficiently extracting and analyzing music performance style features has become an important topic in the field of music information processing. This work focuses on the classification of singing styles of female roles in ethnic opera and proposes an Attention-Enhanced 1D Residual Gated Convolutional and Bidirectional Recurrent Neural Network (ARGC-BRNN) model. The model uses a Residual Gated Linear Unit with Squeeze-and-Excitation (RGLU-SE) block to efficiently extract multi-level features of singing styles and combines a Bidirectional Recurrent Neural Network to model temporal dependencies. Finally, it uses an attention mechanism for global feature aggregation and classification. Experiments conducted on a self-constructed dataset of ethnic opera female role singing segments and the publicly available MagnaTagATune dataset show that the classification performance of the ARGC-BRNN model outperforms other comparison models. The model achieves an accuracy of 0.872 on the self-constructed dataset and an Area Under Curve of 0.912 on the MagnaTagATune dataset. The proposed model improves the results by 0.44% and 0.46%, respectively, compared to other models. The model also demonstrates significant advantages in training efficiency. The results indicate that the ARGC-BRNN model can effectively capture music singing style features, providing technical support for the digital and intelligent analysis of ethnic opera art.

Automated dotted arabic expiration date extraction using optimized convolutional autoencoder and custom CRNN

In this research, we presented a novel method for extracting dotted Arabic expiration dates automatically. Our approach utilizes an optimized convolutional autoencoder combined with a bidirectional to reconstruct Arabic dot-matrix expiration dates into their corresponding filled-in formats. To recognize these dates, we customized a lightweight convolutional recurrent neural network. Due to the unavailability of a suitable dataset for Arabic dot-matrix expiration dates, we addressed this gap by generating synthetic images using a custom-designed Arabic dot-matrix True Type Font. Our model was trained on a synthetic dataset comprising 3,287 images, spanning the years 2019 to 2027. We evaluated our model performance in recognizing expiration dates from the reconstructed images. The proposed system achieved a test accuracy of 99.4% on a dataset of 658 images. Additionally, the image reconstruction process attained a Structural Similarity Index score of 0.46. This method can also be applied to other tasks, such as image translation. Furthermore, the pipeline can be integrated into automated manufacturing systems to identify and sort products based on expiration dates, eliminating the inefficiencies of manual data entry. By improving both accuracy and efficiency, this solution offers a practical advancement for recognizing Arabic dot-matrix expiration dates in industrial settings.

A comparative study of deep learning efficiency in the classification of electrical faults of permanent magnet synchronous motor

Abstract Permanent Magnet Synchronous Motors (PMSMs) are widely used in modern industrial applications due to their high efficiency, reliability, and compact size. However, faults in PMSMs, such as stator winding failures, can lead to significant performance degradation and operational failures. Traditional fault detection methods often rely on signal processing and manual analysis, which may be time-consuming and lacking in accuracy. This study explores the application of deep learning techniques for automated fault detection in PMSMs. The deep learning models based on Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) are employed to classify electrical faults in the motor data, which includes the scalogram images of stator current signal allowing models to learn fault patterns. The performance of the used networks has been compared, in order to choose the reliable one for classification purposes and hence to be utilized for developing the prediction system. The experimental results show that the ResNet50 has better capability to classify the variation of data used where it could achieve 100% of accuracy, recall, precision, and F1 score as compared to other techniques.

Prediction of Parkinson’s Disease and Severity of Disease using Machine Learning and Deep Learning

Abstract - Parkinson's Disease (PD) is a progressive neurodegenerative disorder that affects motor functions, speech, and cognition. This project aims to develop an intelligent system for the detection of Parkinson’s Disease using both Machine Learning (ML) and Deep Learning (DL) techniques. We utilize publicly available datasets containing biomedical voice measurements and other biometric signals. The proposed approach involves data preprocessing, feature extraction, and model training using algorithms such as Support Vector Machines (SVM), Random Forest, and deep neural networks like Convolutional Neural Networks (CNN) or Recurrent Neural Networks (RNN). Our results demonstrate that deep learning models, especially those tailored for time-series or audio data, offer improved accuracy compared to traditional ML models. This project highlights the potential of AI-based tools in assisting healthcare professionals with early and non-invasive diagnosis of Parkinson’s Disease.

ResCapsnet: a capsule network with CRAM and BiGRU for sound event detection

Sound event detection (SED) is a challenging task where ambient sound events are detected from a given audio signal, which includes categorizing the events and estimating their onset and offset times. Deep learning methods such as convolutional neural networks (CNN) and recurrent neural networks (RNN) have achieved promising performance in SED. However, for overlapping sound events, existing deep learning methods are still limited in detecting individual sound events from their mixtures. Inspired by the success of the dynamic routing mechanism of the capsule network (CapsNet), this paper proposes a capsule network model (ResCapsnet-BiGRU) based on a customized residual attention module (CRAM) and bidirectional gated recurrent unit (BiGRU). CRAM is utilized to extract features from log-mel spectrograms that are relevant to sound events. Through dynamic routing, the capsule network can address the overlapping sound events problem. In addition, the BiGRU with time-distributed fully connected layers is adopted to obtain contextual information. Our proposed method was evaluated on two datasets: the Vehicle Weakly Labeled Sound Dataset (VWLSD , DCASE 2017 Task 4) and the Domestic Environment Sound Dataset (DESD , DCASE 2022 Task 4). It achieved F-scores of 62.1% and 75.9% on the Audio Tagging (AT) task, and 54.1% and 59.0% on the sound event detection (SED) task, respectively. The source codes are available at https://github.com/123sunbing/ResCapsnet.git.

Computer Vision-based Approach Using Deep Learning for Breast Cancer Rehabilitation Evaluation: A Comparative Performance of CNN and RNN Using Skeleton Data

Breast cancer rehabilitation plays a crucial role in the recovery process of post-treatment, emphasizing the significance of effective evaluation systems for rehabilitation exercises. This study explores into the utilization of depth-sensing technologies, particularly focusing on skeleton data, in assessing the efficacy of these exercises. Leveraging the ability of deep learning techniques, specifically Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), this study aims to compare their performance in evaluating breast cancer rehabilitation exercises based on skeleton data. The study conducts a comprehensive regression analysis to assess and compare the models' capabilities. The experimental results reveal insights into the comparative effectiveness of CNN and RNN in evaluating the nuances of these exercises, shedding light on their potential applications in enhancing breast cancer rehabilitation evaluation systems.

Sequence modeling for predicting three-dimensional plasma etching profiles with deep learning

Three-dimensional plasma etching profile evolution remains a fundamental challenge in semiconductor process engineering, especially given its complexity and computational cost. In this work, we present a data-driven sequence modeling framework for efficient and accurate prediction of three-dimensional plasma etch profiles. A physically calibrated simulation model was used to generate a comprehensive dataset, encompassing over 2000 time-resolved vertical (y–z) profile sequences and 1000 depth-resolved lateral (x–y) pattern clips, covering a wide range of process conditions and layout geometries representative of modern optical proximity correction patterns. Building on this dataset, we develop a Conditional Convolutional Recurrent Neural Network designed to predict the temporal and spatial evolution of etching contours. The model achieves robust predictive accuracy, with an average Structural Similarity Index Measure above 0.94 across predicted frames, and demonstrates consistent performance for both simple and complex feature arrangements. The sequence modeling approach allows for rapid generation of full etch profile sequences, reducing computation time by orders of magnitude compared to traditional simulation workflows. This framework provides an efficient solution for data-driven plasma etch profile prediction, supporting potential applications in process monitoring, virtual metrology, and layout-level risk assessment.

A Multi-Semantic Feature Fusion Method for Complex Address Matching of Chinese Addresses

Accurate address matching is crucial for the analysis, integration, and intelligent management of urban geospatial data and is also a key step in achieving geocoding. However, due to the complexity, diversity, and irregularity of address expression, address matching becomes a challenging task. This paper proposes a multi-semantic feature fusion method for complex address matching of Chinese addresses that formulates address matching as a classification task that directly predicts whether two addresses refer to the same location, without relying on predefined similarity thresholds. First, the address is resolved into address elements, and the Word2vec model is trained to generate word vector representations using these address elements. Then, multi-semantic features of the addresses are extracted using a Text Recurrent Convolutional Neural Network (Text-RCNN) and a Graph Attention Network (GAT). Finally, the Enhanced Sequential Inference Model (ESIM) is used to perform both local inference and inference composition on the multi-semantic features of the addresses to achieve accurate matching of addresses. Experiments were conducted using Points of Interest (POI) address data from Baidu Maps, Tencent Maps, and Amap within the Chengdu area. The results demonstrate that the proposed method outperforms existing address matching methods, with precision, recall, and F1 values all exceeding 95%. In addition, transfer experiments using datasets from five other cities including Beijing, Shanghai, Xi’an, Guangzhou, and Wuhan show that the model maintains strong generalization ability, achieving F1 values above 84% in cities such as Xi’an and Wuhan.

ARTIFICIAL INTELLIGENCE IN CYBERSECURITY: THREATS, DEFENSES, AND FUTURE DIRECTIONS

This study explores the usage of Artificial Intelligence and machine learning technology in modern cybersecurity. When cyberattacks becoming more frequent, organizations are turning to AI-powered solutions to improve their defense systems. Adoption of AI technologies like machine learning and deep learning is shown to improve the accuracy of threat detection, but data reliability, model interpretability, and bias in decision-making systems remain significant issues. Research has proven AI to be most successful in preventing large-scale attacks, i.e., DDoS, through the recognition of patterns in network traffic based on models like Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs). The application of AI technologies, such as machine learning and deep learning, has been shown to improve the speed and precision of threat detection so that it becomes easier to identify and counteract previously unknown vulnerabilities. Despite these advantages, very low rates of organizations have fully implemented AI-driven security solutions, and confidence in AI decision-making still affects deployment. This article also highlights increasing incidents of AI-driven cyber attacks, such as AI-created malware, automated phishing, and deepfake identity theft, that are outrunning traditional security controls. This analysis also emphasizes the growing importance of application of AI in cybersecurity with underscoring its potential to strengthen defens systems against zero-day attacks and similar evolving threats. However, it also reviews some of the challenges that need to be addressed for more effective integration of AI models in cybersecurity, such as the reliability of the data that is being used for its' training. With the development in AI technologies, machine learning based models are expected to play an increasingly crucial role in protection of both businesses and individuals from sophisticated cyberattacks. And despite all the advantages, the full implementation of AI-driven security solutions remains low. What is more, concerns about the trustworthiness of AI decisions and the lack of transparency in AI models still remains.

Deep Learning for Video Summarization

Abstract This project aims to develop a deep learning-based video summarization system that utilizes Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) to analyze video content and generate concise summaries. The system will automatically identify key objects, events, and scenes in videos, and create summaries that capture the essential information. The project will explore various deep learning architectures and techniques to improve the quality and efficiency of video summarization.

Evolution of Machine Learning: A Foundation for Intelligent Systems

Machine learning has transformed credit card fraud prevention by enabling more sophisticated detection capabilities compared to traditional rule-based systems. The evolution began with supervised learning algorithms like logistic regression, decision trees, and random forests, which classified transactions based on historical data patterns. Unsupervised learning techniques, including clustering algorithms and autoencoders, emerged as vital tools for detecting previously unknown fraud patterns without labeled data. Deep learning architectures such as Recurrent Neural Networks and Convolutional Neural Networks have further revolutionized transaction monitoring by processing sequential data and recognizing complex patterns across multiple dimensions. These advanced technologies can detect sophisticated fraud schemes that would remain invisible to conventional methods. Future directions include hybrid architectures combining multiple model types, federated learning for privacy-conscious collaborative training, and adversarial techniques to enhance resilience against emerging threats. Challenges persist in balancing detection accuracy with false positive rates, meeting regulatory requirements for transparency and fairness, and developing adaptive systems capable of responding to continuously evolving fraud tactics without complete redesign.

Infant Cry Scene Detection and Classification Using High-Frequency Wavelet Scattering Transform and Convolutional Recurrent Neural Network

Infant Cry Scene Detection and Classification Using High-Frequency Wavelet Scattering Transform and Convolutional Recurrent Neural Network

Chaotic crow search enhanced CRNN: a next-gen approach for IoT botnet attack detection

Internet of things (IoT) botnet attack detection is crucial for reducing and identifying hostile threats in networks. To create efficient threat detection systems, deep learning (DL) and machine learning (ML) are currently being used in many sectors, mostly in information security. The botnet attack categorization problem is difficult as data dimensionality increases. By combining convolutional and recurrent neural layers, our work effectively addressed the vanishing and expanding gradient difficulties, improving the ability to capture spatial and temporal connections. The problem of weight decaying and class imbalance affects the accuracy rate of the existing DL models. In convolutional neural network (CNN), fully connected layer optimizes the hyperparameters by utilizing its comprehension of the chaotic crow search method. The chaotic mapping maintains equilibrium between the global and local search spaces. The crow's strategy for hiding food is the main source of inspiration for optimizing the learning rate, weight, and bias components involved in the prediction process. When compared to other existing algorithms, the UNSW-NB15 dataset's results for IoT botnet attack detection in the presence of a high degree of class imbalance demonstrated the effectiveness of the proposed convolutional recurrent neural network (CRNN) boosted with chaotic crow searching algorithm, which produced the highest detection rate with the lowest false alarm rate.

Speech signal-based accurate neurological disorders detection using convolutional neural network and recurrent neural network based deep network

Speech signal-based accurate neurological disorders detection using convolutional neural network and recurrent neural network based deep network

AIoT Fault Detection for Firefighting Pump Maintenance Services Based Metaheuristics and Combined Deep Learning Methodologies

ABSTRACTFirefighting pumps are vital components in fire safety systems, and their proper maintenance is essential for operational reliability. Conventional maintenance methods significantly depend on manual inspection and labor‐intensive procedures, which are time‐consuming and require significant personnel and capital expenses, particularly in large infrastructures. This paper introduces a novel fault detection framework leveraging artificial intelligence of things (AIoT) technology to enhance firefighting pump maintenance services. An advanced hybrid deep learning approach, IPSO‐GRU‐CNN, is developed to improve failure classification accuracy. The improved particle swarm optimization (IPSO) methodology is employed for hyperparameter optimization of the gated recurrent unit and convolutional neural network (GRU‐CNN) model, demonstrating superior performance to conventional optimization methods such as PSO and random search. The IPSO‐GRU‐CNN model is extensively compared with various deep learning architectures, including recurrent neural networks (RNN), CNN, long short‐term memory (LSTM), GRU, and CNN‐GRU, to assess its classification accuracy and efficiency. The suggested AIoT framework optimizes the fault detection process and demonstrates a practical and scalable solution for industrial applications, significantly reducing labor costs and capital expenses associated with the maintenance services of firefighting pumps. Experimental results demonstrated that the developed framework outperforms conventional techniques in terms of classification accuracy and error. Comparing across conventional techniques, IPSO‐GRU‐CNNs acquire the most significant enhancements of 73.37% loss, 98.88% validating loss, 25.84% CP, 89.72% validating CP, 74.64% MAE, 97.36% validating MAE, 74.21% MSE, 99.9% validating MSE, 5.8% PRE, 5.78% validating PRE, 5.06% REC, and 5.2% validating REC. This framework offers a robust and efficient solution for predictive maintenance in firefighting pump systems, facilitating early fault detection and reducing downtime.

Solving spatial-temporal PDEs with arbitrary boundary conditions using physics-constrained convolutional recurrent neural networks

Solving spatial-temporal PDEs with arbitrary boundary conditions using physics-constrained convolutional recurrent neural networks

Optoelectronic Pipeline Architecture of Convolutional RNN for Energy Efficient Inference at the Speed of Light

Optoelectronic Pipeline Architecture of Convolutional RNN for Energy Efficient Inference at the Speed of Light