Search

Deep Neural Networks (DNNs) have revolutionized various fields through their ability to model complex patterns, yet their performance critically hinges on the availability of large-scale, accurately labeled datasets. The costly and time-consuming process of creating such datasets has motivated research into Learning with Noisy Labels (LNL), which aims to reduce reliance on perfect labels. This paper introduces CoFix, an innovative LNL framework that integrates sample selection with semi-supervised learning techniques to address the challenge of noisy labels in training. CoFix employs a Gaussian Mixture Model (GMM) to dynamically segment the training data into clean and noisy subsets, leveraging semi-supervised learning approaches on both. Inspired by the FixMatch algorithm, CoFix refines consistent regularization and pseudo-labeling strategies, enhancing augmentation strategies and temperature sharpening techniques. Additionally, CoFix explores label smoothing to augment the loss function, further refining model performance. Our experiments demonstrate CoFix’s superiority over state-of-the-art methods, achieving significant improvements in fewer training epochs, particularly in lower noise scenarios. The robustness and versatility of CoFix are evident through its consistent performance across various benchmark datasets and noise levels. The contributions of this paper include a novel LNL method with enhanced generalization capability, an investigation into the impact of label smoothing on the loss function, and extensive testing that confirms CoFix’s efficiency and adaptability to different noise levels. [ABSTRACT FROM AUTHOR]

This paper presents a quantitative analysis of the nonlinearities of the positive predictive value (PPV) and its effect in evaluating two-class pattern classification models with imbalanced datasets. The analysis is made through an expression of the PPV as a function of two other classification ratios that are invariant to the data imbalance —the true positive rate (TPR) and false positive rate (FPR), and σ — the imbalance ratio (IR) of the dataset such that PPV =σTPR/(σTPR+FPR). The curvatures of PPV in the three-dimensional TPR–FPR–σ space are studied using the Hessian matrix, from which a saddle-shaped 3D surface in the space is revealed. This paper explores the nonlinear behaviors of PPV around the critical points, identified at FPR =σTPR on the saddle surface, along with its scaling and sensitivity issues as performance measurements in model evaluation. The effect of the nonlinearities of PPV for the F1 and MCC metrics on imbalanced datasets is also studied. It is warned through the results of this study that the evaluations of classification models could be misleading if without an awareness and understanding of the nonlinearities associated with the PPV and its relevant metrics on imbalanced datasets. [ABSTRACT FROM AUTHOR]

With the development of mobile edge computing and neural network Deep Learning (DL), more and more scholars are studying the combination of the two. This paper mainly studies the application of mobile edge computing and neural network DL in IoT computing collaboration and data-aware routing algorithms. Therefore, this paper proposes the deployment options of MEC technology and ETSIMEC in mobile edge computing, combining mobile edge computing with DL, designs an optimization algorithm based on Markov decision process and feature expression learning, and then analyzes and optimizes IoT computing and VANET routing algorithm. In order to have a clearer direction for the optimization algorithm, this paper also designs the edge computing model training and experiment comparison, the DL algorithm comparison experiment, and the routing algorithm simulation experiment and performance analysis. Combined with the experimental results, it is optimized and compared with traditional IoT computing and routing algorithms. Finally, it is concluded that the computing efficiency of IoT computing based on edge computing and DL designed in this paper is 21.33% higher than that of traditional IoT computing. The efficiency of the routing algorithm based on edge computing and DL designed in this paper is 9.29% higher than that of the traditional routing algorithm. [ABSTRACT FROM AUTHOR]

Video anomaly detection has always been a challenging task in computer vision due to data imbalance and susceptibility to scene variations such as lighting and occlusions. In response to this challenge, this paper proposes an unsupervised video anomaly detection method based on an attention-enhanced memory network. The method utilizes a dual-stream network structure of autoencoders, enhancing the model's learning ability for important features in appearance and motion by introducing coordinate attention mechanisms and variance attention mechanisms, emphasizing significant characteristics of static objects and rapidly moving regions. By adding memory modules to both the appearance and motion branches, the network structure's memory information is reinforced, enabling it to capture long-term spatiotemporal dependencies in videos and thereby improving the accuracy of anomaly detection. Furthermore, by optimizing the network structure's activation functions to handle negative inputs, it enhances its nonlinear modeling capabilities, enabling better adaptation to complex environments, including variations in lighting and occlusions, further improving the effectiveness of anomaly detection. The paper conducts comparative experiments and ablation studies using three public available datasets and various models. The results demonstrate that compared to baseline models, the AUC performance is improved by 3.9%, 4.7%, and 1.7% on UCSD Ped2, CHUK Avenue, and ShanghaiTech datasets, respectively. When compared with the other models, the average AUC performance is improved by 4.3%, 5.4%, and 6.2%, with an average improvement of 8.75% in the ERR metric, validating the effectiveness and adaptability of the proposed method. The code can be obtained at the following URL: https://github.com/AcademicWhite/AEMNet. [ABSTRACT FROM AUTHOR]

With a rapid development of artificial intelligence technology, fine-grained image classification has gained widespread application. For mobile terminals, this paper introduces an image classification method built on MobileViT, and it can apply into fine-grained image classification. The original MobileViT model has been optimized in three ways. Initially, the h-swish activation function is used to enhance the network performance. Second, the cross-entropy loss function is used to further realize the parameter optimization and model accuracy improvement. Finally, a dropout layer is joined before the fully connected layer can effectively decrease the model recognition time and prevent over-fitting. Experimental data on public tomato disease datasets demonstrate that the improved fine-grained image classification method put forward in this paper exhibits higher classification accuracy, better stability and network generalization ability than other models. [ABSTRACT FROM AUTHOR]

The rapid growth of contents on the Web in different languages increases the demand of Cross-Lingual Information Retrieval (CLIR). The accuracy of result suffers due to many problems such as ambiguity and drift issue in query. Query Expansion (QE) offers reliable solution for obtaining suitable documents for user queries. In this paper, we proposed an architecture for Hindi–English CLIR system using QE for improving the relevancy of retrieved results. In this architecture, for the addition of term(s) at appropriate position(s), we proposed a location-based algorithm to resolve the drift query issue in QE. User queries in Hindi language have been translated into document language (i.e. English) and the accuracy of translation is improved using Back-Translation. Google search has been performed and the retrieved documents are ranked using Okapi BM25 to arrange the documents in the order of decreasing relevancy to select the most suitable terms for QE. We used term selection value (TSV) for QE and for retrieving the terms, we created three test collections namely the (i) description and narration of the Forum for Information Retrieval Evaluation (FIRE) dataset, (ii) Snippets of retrieved documents against each query and (iii) Nearest-Neighborhood (NN) words against each query word among the ranked documents. To evaluate the system, 50 queries of Hindi language are selected from the FIRE-2012 dataset. In this paper, we performed two experiments: (i) impact of the proposed location-based algorithm on the proposed architecture of CLIR; and (ii) analysis of QE using three datasets, i.e. FIRE, NN and Snippets. In the first case, result shows that the relevancy of Hindi–English CLIR is improved by performing QE using the location-based algorithm and a 12% of improvement is achieved as compared to the results of QE obtained without applying the location-based algorithm. In the second case, the location-based algorithm is applied on three datasets. The Mean Average Precision (MAP) values of retrieved documents after QE are 0.5379 (NN), 0.6018 (FIRE) and 0.6406 (Snippets) for the three test collections, whereas the MAP before QE is 0.37102. This clearly shows the significant improvement of retrieved results for all three test collections. Among the three test collections, QE has been found most effective along with Snippets as indicated by the results with the improvements of 6.48% and 19.12% over FIRE and NN test collections, respectively. [ABSTRACT FROM AUTHOR]

This paper proposes an image compression algorithm based on Swin Transformer and residual network (STRN), aiming to reduce blurring and distortions in traditionally compressed images. The algorithm utilizes a dual-channel mechanism to remove artifacts from the image, which takes advantage of the complementary features of the transform and residual networks. The Swin Transformer networks address the issue of long-range dependency, leading to an enhanced and improved reconstructed image quality. The residual network is an effective network that mitigates gradient loss and recovers image details during the image compression process. The paper demonstrates that image compression can be achieved by training a convolutional network based on a transformer and residuals network, which significantly reduces artifacts and provides better reconstructed image quality compared to previously used and current mainstream methods based on traditional convolutional neural networks. The proposed approach can remove blocking artifacts by subtracting estimated artifacts from the input image, while still preserving most of the original details. Therefore, our proposed method is highly effective in improving image quality and reducing visual artifacts caused by traditional compression methods. Moreover, this method is useful for enhancing image transmission and storage efficiency in various computer vision systems that employ digital visual codecs. [ABSTRACT FROM AUTHOR]

Glaucoma is a major cause of irreversible blindness caused by optic nerve damage. The ophthalmologist uses retinal examination of the dilated pupil to diagnose this disease. Since this diagnosis is a manual and laborious procedure, an automated technique is required for faster diagnosis. Automated retinal image processing is deemed a competitive research field owing to its lower accuracy results, complication and improper effects related with it. Therefore, Optimized Improved Random Forest fostered Glaucoma Detection from Fundus Retinal Images (IRF-MOSOA-GD) is proposed in this paper. Here, Images are acquired through the datasets of DRISHTI-GS, ORIGA and RIM_ONE and given to the pre-processing. The pre-processing is carried out utilizing the Savitzky–Golay Denoising technique for eliminating the noise at the input images. Then the pre-processed image is given to the feature extraction phase. In the feature extraction phase, the region features are extracted with the help of the Fuzzy color and Texture histogram (FCTH), Edge histogram and Pyramid Histograms of Orientation Gradients (PHOG) method. Then, the extracted feature is fed to the Improved Random Forest (IRF) classifier for categorizing the normal and Glaucoma images. The hyperparameter of the IRF classifier is tuned with a Multi-Objective Squirrel Optimization Algorithm (MOSOA) to attain better categorization of normal and glaucoma images. The proposed technique is implemented in Java and its efficiency is analyzed under some metrics, like accuracy, F-scores and computational time. The IRF-MOSOA-GD method attains higher accuracy in the DRISHTI-GS dataset at 23.6%, 27.55% and 24.98% higher accuracy compared with existing techniques. [ABSTRACT FROM AUTHOR]

Extracting information from scanned business documents, while a necessary commercial task, continues to be mostly done manually, requiring significant human effort. Current solutions for automated document information extraction still have limited capabilities in regards to user-required customizability and extraction of dataset-specific information, leaving the area as a very active field of research. In this paper, we propose modifications and improvements to our previously developed custom pipeline for extracting and tabulating key-value pairs from commercial invoice documents. Our design changes and additions adapt the pipeline to a wider variety of document types and use cases, primarily through the implementation of dataset-specific configuration files that promote customizability along with new technical modules that address both general and dataset-specific complexities. We compare our pipeline’s performance against current machine learning and commercial solutions on a real-world dataset, and demonstrate that it is able to extract a wider variety of fields while maintaining competitive or greater accuracies compared to the alternate solutions. [ABSTRACT FROM AUTHOR]

In medical imaging, accurate diagnosis heavily relies on effective image enhancement techniques, particularly for X-ray images. Existing methods often suffer from various challenges such as sacrificing global image characteristics over local image characteristics or vice versa. In this paper, we present a novel approach, called G-CLAHE (Global-Contrast Limited Adaptive Histogram Equalization), which perfectly suits medical imaging with a focus on X-rays. This method adapts from Global Histogram Equalization (GHE) and Contrast Limited Adaptive Histogram Equalization (CLAHE) to take both advantages and avoid weakness to preserve local and global characteristics. Experimental results show that it can significantly improve current state-of-the-art algorithms to effectively address their limitations and enhance the contrast and quality of X-ray images for diagnostic accuracy. [ABSTRACT FROM AUTHOR]

Underwater bubble plume images contain a wealth of information on wave field and flow characteristics, which can provide valuable research data for marine development, environmental protection, and underwater surveys. However, based on fusing image features and wave field environment features, identifying accurately the underwater bubble plume is still very difficult. In order to improve the accuracy and robustness of bubble plume identification in complex underwater environments, an underwater bubble plume recognition algorithm based on multi-feature fusion understanding is proposed. In this paper, a weight-independent dual-channel residual convolutional neural network (CNN) for feature extraction of the original optical images and the nonsubsampled contourlet transform (NSCT) low-frequency images, and the multi-scale composite feature map groups are generated. Then adaptive fusion is performed based on the feature contribution of the target in different types of images. Next, logical region of interest (ROI) masks are generated by the attention mechanism and superimposed on the fused image to further highlight the target features. Finally, the multi-scale dual-channel fused feature maps containing ROI masks are used for underwater bubble plume target recognition. The experimental results show that the designed recognition network can effectively fuse the features of the original optical images and the NSCT low-frequency imagers, improve the depth of information fusion, and retain the target texture features and the morphological features while reducing the interference of the background information, and have good recognition accuracy and robustness for multi-scale bubble targets in the underwater environment. [ABSTRACT FROM AUTHOR]

The accurate health status evaluation of lithium-ion batteries is crucial for preemptive identification of potential battery failures and averting hazardous incidents, given its essential role in indicating the extent of battery degradation. The challenge in determining the State of Health (SOH) arises from the absence of a precise and standardized definition, as well as the difficulty in measuring essential input variables. Therefore, this paper utilizes current and voltage data during the charge and discharge process as direct inputs for SOH estimation and proposes a deep learning-based lithium-ion battery SOH estimation approach. Specifically, it leverages Bayesian optimized Convolutional Neural Network (CNN) within a data-driven framework. Experimental results demonstrate that the proposed deep learning method achieves a Mean Absolute Error (MAE) of 1% and a Maximum Error (MAX) below 4% in estimation accuracy, highlighting its enhanced precision and robustness. [ABSTRACT FROM AUTHOR]

Printed circuit boards (PCBs) play a critical role in electronic products. Ensuring these products’ long-term reliability and consistent performance requires effective PCB defect detection. Although existing deep learning models for PCB defect detection are not highly accurate, they often neglect capability considerations. This paper introduces a precise, fast, and lightweight defect detection model, CCG-YOLO, based on an enhanced YOLOv5 model to address this issue. The enhancements in CCG-YOLO can be summarized as follows: (1) Improved Backbone network: The feature extraction ability of the Backbone network is enhanced by introducing a C3HB module, which fosters spatial interaction capabilities. (2) Lightweight feature fusion network: A lightweight convolution structure called Ghost-Shuffle Convolution is incorporated in the feature fusion network, remarkably reducing model parameters while maintaining performance. (3) Efficient residual networking: To enhance model performance further, a CNeB module is introduced based on the ConvNeXt network, which replaces the C3 module in the Neck. CNeB improves model detection accuracy and reduces the number of model parameters. The combination of these enhancements results in impressive performance. CCG-YOLO achieves mean average precision (mAP@0.5) of 99.5% and 88.75% in mAP@0.5:0.95 on the TDD-Net public dataset. Compared with the original YOLOv5s algorithm, CCG-YOLO offers a 4.24% improvement in mAP@0.5:0.95, a 1MB reduction in model size, a 0.472M decrease in the number of parameters, a 0.6G floating point operation reduction in computational complexity, and a 120 frames per second real-time inference speed. These experimental results underscore that the proposed model excels in accuracy and speed and has a compact size for PCB defect detection. Moreover, CCG-YOLO is easily deployable on low-end devices, making it well-suited for meeting the real-time requirements of industrial defect detection. [ABSTRACT FROM AUTHOR]

In order to reduce the probability of leakage and improve detection accuracy, multiple radar equipment is required to perform multiple detections on the same airspace. Due to the overlap of airspace, there are multiple local tracks reported by various radar equipment that belong to the same target. The local tracks reported by each radar require track fusion to form a system track, and track association is a prerequisite for track fusion. This paper proposes a multi-target and multi-sensor track association analysis method, which can distinguish whether local tracks come from the same target and associate local tracks with the same target, providing conditions for track fusion to form system tracks. This method divides track association into two steps: Coarse association and fine association, which greatly improves the accuracy of track association. [ABSTRACT FROM AUTHOR]

The dual-spectrum triocular camera system composed of a binocular visible light camera and mid-infrared camera is used for high-precision thermal fault detection and thermal field reconstruction of industrial equipment. The realization of its function depends on the high-precision camera parameter calibration. The difficulty lies in how to realize the infrared camera calibration quickly and improve the parameter accuracy of multi-lens camera. In this paper, according to the characteristics of the dual-spectral triocular camera system, the theoretical model is constructed, and the circular asymmetric calibration plate under infrared supplementary light is selected as the calibration object through experiments. A method for combining infrared image adaptive histogram enhancement and binarization processing based on the Sauvola algorithm is proposed to effectively calibrate the infrared camera. A global parameter optimization method based on traditional passive vision binocular stereo calibration is proposed. The optimization of experimental parameters finds the reprojection error value is 0.16, which meets the demand of high-precision calibration and solves the problem of difficult-to-calibrate weak image texture of infrared camera and the calibration accuracy of the camera under different imaging capabilities. [ABSTRACT FROM AUTHOR]

This paper presents a new approach for calculating the Euler characteristic in 2D binary images. The problem is addressed using the Three OrThogonal symbol chain code (3OT code), using only one symbol for the calculation of the Euler characteristic. Using this code, it is possible to introduce new geometric concepts represented by the same symbol of the 3OT alphabet and to simplify the overall equation of the Euler characteristic. This process is supported by the proof of a set of theorems and their numerical validation, using a set of binary images with a variable number of holes. Thus, this research proves that the 3OT code can be used not only for image compression as reported in the literature, but also to simplify the expression of the Euler characteristic as well as for the analysis and simplification of the shape of contours. [ABSTRACT FROM AUTHOR]

Graph clustering is an essential task in data analysis. Recently, there has been a notable trend in the application of deep learning in graph clustering. However, it is worth noting that existing deep graph clustering methods primarily focus on the topological information of nodes while overlooking the use of attribute information in the learned feature representations. In addition, they do not address the consistency of the space distribution in attribute and structure clustering results. To tackle these critical issues, a novel structured deep graph clustering network with consistency constraint (CC-DGC) is proposed. The network first constructs an autoencoder to learn and transmit hierarchical attribute information to the graph autoencoder (GAE). Subsequently, the GAE integrates the received hierarchical attribute information with extracted topological information to generate enhanced clustering representations. Moreover, this paper designs a consistency constraint module to promote consistency between the autoencoder and the GAE by optimizing the cluster space distributions they produce. Finally, the feature extraction and clustering classification processes are synchronized and optimized in a self-supervised manner within a unified framework. Extensive experiments illustrate that the proposed CC-DGC demonstrates superiority over the state-of-the-art deep graph clustering methods on five benchmark datasets. [ABSTRACT FROM AUTHOR]

Usually, digital cameras comprise sensor arrays enclosed by Color Filter Arrays (CFAs), mosaics of minute color filters. Thus, every pixel sensor usually records limited spectral data regarding relevant pixels. Demosaicing is defined as the procedure of deducing the misplaced data for every pixel, which plays a vital role in recreating high-quality full-color images. Denoising and demosaicing are the major processes in the camera imaging chain for both videos and images. Here, reconstruction errors occur in these points and have undesirable effects on the final outcome, when it is not appropriately managed. The demosaicing process provokes color and spatial correlation of noises, and it is improved by means of imagining a pipeline. This organized noise usually destroys the quality of the image as well as fails to prevent accurate interpretation of an image. During the mitigation of this structured noise on processed data, denoising techniques diminish the texture and information. Therefore, an effectual demosaicing technique is essential for recreating the full-color image from the defective color samples. Thus, in this paper, an effectual video demosaicing model is proposed using an optimized deep learning system. The designed video demosaicing system achieved better performance with a Peak Signal-to-Noise Ratio (PSNR) of 59.74 dB, Second Derivative, like Measure of Enhancement (SDME) of 63.51, and Root Mean Squared Error (RMSE) of 0.3660. [ABSTRACT FROM AUTHOR]

The compressor is the most core component of refrigeration products, and its performance directly affects the overall performance of the product. The exhaust silencing chamber and coil are key components of a refrigeration compressor, and their working stability and lifespan determine the reliability and economy of the refrigeration compressor. This paper obtains reference data for the fluid domain of the exhaust silencing chamber through simulation analysis and image processing of the existing exhaust silencing chamber structure and inner diameter size of the coil. Then, the optimal model of the prioritized exhaust chamber is obtained through pattern recognition method. Based on the image processing method, the internal flow field of the exhaust chamber are obtained, providing a strong reference for the optimal structural design of the exhaust silencing chamber and coil of the refrigeration compression body. [ABSTRACT FROM AUTHOR]

Lightning protection near the fan is the key point of lightning protection in wind farm, which affects the safe and effective delivery of wind power resources. At present, the three-phase multi-gap arrester installed on a base tower behind the collector can cooperate with the zinc oxide arrester to reduce the system failure caused by the arrester failure. However, the chopped wave generated by the multi-gap arrester when it acts is transmitted to the fan box transformer along the line, which may threaten the insulation of the transformer winding. Based on the above problems, this paper builds a simulation model on ATP to analyze the impact of wave clipping of multi-gap arrester on the insulation of fan transformer windings, considers the conventional installation mode and the limit installation mode of multi-gap arrester, and calculates that the wave clipping overvoltage generated by multi-gap arrester after being struck by lightning in two cases is the impact on the insulation of fan transformer. Through the simulation analysis, it is found that the higher the amplitude of the lightning current, the larger the amplitude of the overvoltage generated by the truncated wave, the steeper the wave head, and the greater the threat to the insulation of the fan box. The damage degree of round lightning is less than that of counter lightning. When a multi-gap arrester is installed in a collector tower, the 100kA lightning current may destroy the insulation of the fan box transformer winding and require additional protective measures. It is recommended that the multi-gap arrester be installed from the next pole tower of the collector tower in the wind farm. [ABSTRACT FROM AUTHOR]

The development of reduced-order models has been a pivotal advancement in the computational analysis of fluid dynamics, substantially simplifying the complexity and boosting the efficiency of simulations. The accuracy and practicality of these models largely depend on the reduction techniques applied and the inherent characteristics of the fluid dynamics systems they represent. In this paper, we introduce an innovative machine-learning framework for assessing model uncertainty in computationally intensive reduced-order models. By combining subspace construction methods with advanced Bayesian inference techniques, our approach effectively captures the posterior distribution of model parameters, thereby providing an accurate representation of uncertainty in aerodynamic performance predictions. We employ the NACA0012 airfoil as a case study to validate our method’s ability to enhance the efficiency of reduced-order models and precisely measure the uncertainty inherent in predictions made by recurrent neural networks. It is important to note that our approach is influenced by specific constraints and variables that significantly impact the mean and variability of the predicted final lift coefficient distribution. Our findings indicate that setting the goodness of fit (R2) threshold above 0.985 markedly improves the correlation between Computational Fluid Dynamics (CFD) outcomes and model predictions, increasing from 72.2% to 97.9% as the interval amplification factor adjusts from 1.5 to 3. However, this adjustment causes a considerable expansion of the confidence interval, from 0.0737 to 0.1282, an increase of over 70%. Despite these challenges, our machine learning-based methodology provides essential insights into the further development of reduced-order modeling and uncertainty quantification in fluid dynamics. This highlights the need for ongoing research into the model parameters, especially in applications concerning aircraft control systems, to meet design specifications and ensure system reliability. [ABSTRACT FROM AUTHOR]

Multi-interest sequential recommendations leverage users' historical behavior to provide recommendations that match multiple interests. Most of these methods have not fully extracted higher-order information hidden in users' interactions and have overlooked the multiple features of items. To this end, this paper proposes a multi-interest model called "multi-interest sequential recommendation with simplified graph convolution and item multi-features (SGCMF)". Firstly, a simplified graph convolution module is designed based on bipartite graphs, which utilizes mean pooling to aggregate neighboring information and employs a feedforward neural network (FNN) for nonlinear transformations and combinations. This method reduces redundant information and captures higher-order relationships, thereby simplifying the complexity of modeling high-order interactions and improving prediction accuracy. Secondly, an item multi-feature extraction module is proposed, which represents item features with multiple vectors, and analyzes each feature from multiple perspectives while preserving important relationships between features. The model correlates multiple features of the item with user interests, thereby achieving a fine-grained analysis of user interests. Extensive experiments are conducted on five real-world scenarios, and the results are compared with state-of-the-art methods. The experimental results show that SGCMF outperforms other baselines. [ABSTRACT FROM AUTHOR]

Deep learning classification is commonly used to assess biomedical data. Fetal heart rate (FHR) signal data monitor the health of both the mother and the fetus and prevent mobility and death in fetuses at risk of hypoxia. But, the data imbalance is a major problem and happens frequently. To overcome this problem, a Deep Siamese domain adaptation convolutional Neural Network optimized with Honey Badger Algorithm (HBA) is proposed in this paper for Fetal Arrhythmia Detection and Classification from ECG signals (DSDACNN-HBA-FAC). The input ECG signals are collected from the fetal ECG dataset. Then, the ECG data are pre-processed using coherence shock filtering (CSF), which eliminates artifacts, like baseline drift, contact noise, muscle artifacts, electrode motion, electromyography artifacts. Using the improved non-subs sampled Shearlet transform (INSST) method, the pre-processed ECG data recover the mean, standard deviation, root mean square, skewness, kurtosis, zero-crossing, autoregressive coefficients, estimated entropy, and Higuchi's fractal dimension. The DSDACNN classifier utilizes the features extracted to categorize the output as normal and arrhythmia. In general, DSDACNN does not agree on any optimization techniques to define the optimum parameters and to ensure accurate classification. Therefore, HBA is employed to optimize the DSDACNN weight parameters. The proposed method is implemented in MATLAB. The performance metrics, like sensitivity, precision, recall, f-measure, specificity, accuracy, computation time, RoC, and error rate are evaluated. The performance of DSDACNN-HBA-FAC achieves 3.101%, 7.12%, 7.73% high accuracy, 51.136%, 59.04%, 44.51% lower computation Time and 2.292%, 5.365%, 1.551% greater AUC compared with the existing techniques, like Fetal Arrhythmia Detection with Classification from ECG Signals: A NonInvasive Method (ANN-FAC), Semi-supervised active transfer learning for fetal ECG arrhythmia identification (DNN-FAC), fetal arrhythmia detection using adaptive single channel electrocardiogram extraction (CNN-FAC), respectively. [ABSTRACT FROM AUTHOR]

This paper investigates the correlation between a pilot's brain electroencephalographic (EEG) activity and his driving posture, addressing the intricate relationship between cognition and behavior in aviation. We designed and implemented a simulated experiment that recorded the fly pilot's attitude and collected EEG information as experimental data. The experiment is only based on EEG data to predict flight posture (pull, down, left, right). We propose a flight-driving attitude prediction model (CA-FAP) based on CEBRA and self-attention mechanism, with a prediction accuracy of 0.83. This model is better than common spatial pattern (CSP) and uniform manifold approximation and projection (UMAP) dimension reduction methods in the experiment. Moreover, a better effect can be obtained in the larger attitude radian dataset (accuracy is 0.85), and the effect is not obvious in the dataset of closing the six-axis motion platform, indicating that the model prediction of flight attitude is closely related to the pilot position transformation. By comparing the prediction effect of each category, the pull-up and drop are better than the steering prediction result. The study can help pilots adjust their posture and decisions, and serve as a basis for studying flight pilot cognitive load, mental load, mood changes, and flight performance. [ABSTRACT FROM AUTHOR]

Public computer education systems provide students essential opportunities to enhance computer literacy and information skills. However, the widespread adoption of online education technology exposes the field to several critical security risks. Threats, such as malware infections, data breaches, and other network intrusions, are all challenging the security of education systems, posing potential hazards to students' personal information and even the entire teaching environment. To spur further work into specialized anomaly detection techniques for computer education, this paper presents an anomaly detection framework tailored for network services in computer education environments to safeguard these systems. Specifically, the proposed approach learns from large-scale online educational traffic data to classify the security state into five alert levels, enabling more granular anomaly detection and analysis. To assess their detection performance, deep learning and traditional machine learning algorithms are implemented and compared for multi-class intrusion classification. The results show that the proposed framework provides an effective security solution to bolster the integrity and stability of computer education systems against evolving network threats, enhancing threat intelligence to inform proactive security by detecting and characterizing anomalies through multilevel classification. [ABSTRACT FROM AUTHOR]

With the increasing scale of mobile cellular network applications, 5G mobile network infrastructure provides customizable services to users in the form of network slices. How to effectively allocate existing resources in real dynamic networks with time-varying network utility is a key issue that previous work did not consider. This paper first initializes the multi-resource allocation problem of network slicing in an online manner, where the utility function is set to change over time. Therefore, we propose Metis, an online network slicing resource allocation framework that combines the time-varying nature of the network utility function given bandwidth and processing power constraints with the requirement of virtual network function isolation. The goal is to maximize the cumulative network utility in the long term and specify multiple resource allocation problems by utilizing concave optimization methods. In addition, a distributed algorithm based on the online alternating direction method of multipliers with regret optimization has been developed to achieve optimal resource allocation. Our mathematical analysis proves that Metis can provably converge to the optimal solution and the result of experiments demonstrates a steady state behavior of Metis, which converges in dynamic network settings. [ABSTRACT FROM AUTHOR]

Over the past decade, a wide range of brain–computer interface (BCI) applications with artificial intelligence (AI) assistance have emerged. While using a head-mounted electroencephalogram (EEG) device for BCI application, it often requires a user to wear the device close to the scalp so that clear EEG signals can be obtained from the head directly. Those signals can be used to determine a user's mental state. Due to the head size differences between Chinese and European users, as well as the differences between Chinese users themselves, EEG headsets with fixed-electrode-point will enable some electrodes fail to capture a user's EEG data accurately. Combined with previous research and related design methodology, the design of a head-mounted EEG device is proposed in this paper. In this study, Chinese youth are regarded as the target user group. After evaluating different designs of device structures, the possible positions and movement methods of electrodes, an EEG headset with movable electrodes for young Chinese users is finally delivered. [ABSTRACT FROM AUTHOR]

In the last years, endowing the machine with emotional and behavioral intelligence has been one of the most challenging issues in the human–computer interaction area. In this context, this paper investigates the detection of simultaneous fear emotion and deception behavior in speech. To do so, a set of 72 pitch-based features has been investigated first to recognize fear and deception separately. Then, different feature selection techniques have been used in order to select the most relevant ones that best discriminate between fear/nonfear and deception/nondeception classes. Next, a decision-level fusion approach based on the belief theory has been proposed to infer whether fear and deception are detected simultaneously. Simulation results carried on databases dealing with fear/nonfear emotions and deception/truth behaviors have shown classification results reaching 83.33% and 72.45% as accuracy rates for fear and deception classifiers, respectively. The proposed fusion approach has revealed a correspondence between fear emotion and deception behavior in speech modality. [ABSTRACT FROM AUTHOR]

This paper is dedicated to developing a cost-effective ground-moving object detection method in aerial videos. Without limitations on types, quantity, and distribution of moving objects, which were required by the previous methods, the proposed approach can detect various ground-moving objects in aerial videos captured through diverse flying states in various ground appearances. The proposed method is mainly composed of generation of the appropriate feature points, detection of moving objects, and target tracking. In the originality of our work, a novel detection strategy designed for ground-moving objects is based on change detection of Delaunay triangulation (CDDT) and a three-step motion vector search-based tracking algorithm is further exploited for enhancing the detection rate. Experimental results show that our method can achieve the detection rate of at least 95% (roughly similar to the famous existing state-of-the-art methods) and 0.03 second/frame (far less than the famous existing methods) using test videos (containing only several moving objects distributed in a sparse space) in the previous methods compared. Besides, the average detection rate of 86.81%, average false detection rate of 9.44%, and a frame rate of about 33 fps can be obtained using our test videos captured in the complicated ground appearances. This result makes the proposed method more attractive for detecting various ground-moving objects in aerial videos, when compared to other approaches, and can also achieve cost-effective performance. [ABSTRACT FROM AUTHOR]

Petroleum is a critical industrial material. In the process of oilfield development, the Fisher's optimal segmentation method is often used to solve the problem of division and combination of layer series of development. However, when encountering the problem of long samples, this method has particularly obvious drawbacks due to the high storage requirements of the calculation process. Therefore, in practical work, the Fisher's optimal binary segmentation method is generally used instead. Although it avoids storage problems, it is prone to falling into local optima. On the basis of analyzing the shortcomings of Fisher's optimal segmentation and optimal binary segmentation algorithms, this paper processes a genetic clustering algorithm. This algorithm overcomes the problem of Fisher's optimal binary segmentation algorithm easily falling into local optima and solves the problem of high storage capacity requirements in the calculation process of Fisher's optimal segmentation algorithm. Taking the data of 17 subzones in S-2 8-11 sand groups of the Shahejie Formation of the Lower Tertiary in the Dongxin area as an example, this algorithm is applied to divide and combine layer series of development. The experimental results show that the algorithm's partitioning results are reasonable and can optimize the selection of development layers and provide decision support for the production of oil and gas resources. [ABSTRACT FROM AUTHOR]

Recent deep learning-based watermarking methods have achieved impressive results. However, they struggle with unknown distortions and often suffer from poor generalization, slow convergence, unstable training, and degraded visual quality in watermarked images. To address the above problems, this paper proposes a two-stage separable adversarial distortion (TSAD)-based robust watermarking algorithm for diffusion tensor imaging (DTI). The algorithm uses a noise-free end-to-end network in the first stage for learning and training DTI images. In the second stage, it fixes the watermark embedding network trained in the first stage, interacts the noise distortion network with the watermark extraction network to perform adversarial training for improving robustness. Experimental results show that our method achieves comparable or better robustness to seen distortions and better robustness to unseen distortions, along with enhanced stability, faster convergence, and improved visual quality in watermarked DTI images. [ABSTRACT FROM AUTHOR]

Cleaning control based on changes in cleaning loss rate and impurity rate has emerged as a hot topic in the research on intelligent control for rice and wheat combined harvesters. However, numerous operation parameters can lead to deviations beyond the normal range of cleaning loss rate and impurity rate. The impact of cleaning control parameters in rice and wheat combined harvesters on cleaning loss rate and impurity rate often tends to be contradictory. How to combine different contradictory cleaning control strategies to get a more widely used and better cleaning control strategy is a hot issue in the current cleaning control research. In this paper, the granularity of quotient space is introduced into the cleaning control based on operation parameters, and a cleaning control model based on granularity synthesis theory is proposed. This method first constructs a knowledge base for intelligent control of cleaning tailored to the cleaning loss rate and impurity rate using production rule representation, and considers that these control strategies constitute different quotient spaces, and then organizes these quotient spaces according to granularity synthesis theory to get the cleaning control strategy. The experimental results verify that the validity of the cleaning control based on granular computing is better than fuzzy control. [ABSTRACT FROM AUTHOR]

Nanodrones are insect-sized drones that could fly in complex environments and confined spaces, and act as an emerging tool for covert surveillance and intelligence attacks, which would become a potential threat to national security. Radar has the advantage of wide range, all-day, and all-weather detection ability, making it a means of detecting such threat. First, this paper introduces a pertinent multiple-input multiple-output (MIMO) millimeter-wave (MMW) radar system, with the advantages of low cost and high accuracy. It is utilized to detect three targets: nanodrone, small helicopter, and mechanical bird, through which more detailed features can be obtained. Then, the echo data of the three targets are processed and analyzed, and their distinct micro-Doppler characteristics were obtained. Finally, the Radar Transformer target classification network is used to classify and identify the targets. It has been confirmed that desired results could be achieved through the above process. [ABSTRACT FROM AUTHOR]

Collaborative robots face challenges in achieving high accuracy using existing parameter identification methods, especially in applications like dragging, teaching, and collision detection. To address the issue of robot dynamics identification accuracy in scenarios with unknown parameters, this paper introduces a method for robot joint parameter identification based on the Cuckoo Search-backpropagation neural network (CS-BPNN). Initially, a kinetic model incorporating Coulomb viscous friction is presented. Subsequently, a BPNN is employed to approximate and fit the nonlinear function, while the CS algorithm is integrated into the BPNN model to optimize its parameters. Collected data undergo third-order Butterworth filtering, and a loss function compares predicted values with actual values to ascertain the accuracy of the proposed method. Experimental results demonstrate that the CS-BPNN approach proposed herein can converge the mean square error (MSE) to 9. 7 9 × 1 0 − 4 N⋅m when robot dynamics parameters are unknown. This method boasts of a lower MSE and superior discrimination accuracy compared to traditional methods like least squares (LS), genetic algorithm (GA), and BPNN. Consequently, the method presented in this study not only enhances the accuracy of parameter identification but also offers a fresh perspective for the parameter identification of robot joint dynamics models. [ABSTRACT FROM AUTHOR]

Thanks to the rapid development of deep learning in recent years, image inpainting has made significant progress. As a fundamental task in the field of computer vision, many researchers are committed to exploring more efficient methods, and state-of-the-art research results prove that generative adversarial networks (GAN) have superior performance. However, due to the inherent ill-posedness of image inpainting tasks, these approaches suffer from lack of detailed information, local structural fractures or boundary artifacts. In this paper, we leverage the properties of GAN architecture to process images in more detail and more comprehensively. A novel dual U-Net GAN is designed to inpaint images, which is composed of a U-Net based generator and a U-Net-based discriminator. The former captures semantic information of different scales layer by layer and decodes it back to the original size to repair damaged images, while the latter optimizes the network by combining reconstruction loss, adversarial loss, perceptual loss and style loss. In particular, the U-Net-based discriminator allows per-pixel detail and global feedback to be provided to the generator, guaranteeing the global consistency of the inpainted image and the realism of local shapes and textures. Extensive experiments demonstrate that for different proportions of damage, the images inpainted by our proposed model have reasonable texture structure and contextual semantic information. Furthermore, the proposed model outperforms state-of-the-art models in both qualitative and quantitative comparisons. The code will be available at https://github.com/yjjswu. [ABSTRACT FROM AUTHOR]

Intelligent control has become an important research direction of a combine harvester. However, the impact of cleaning control parameters in rice–wheat combine harvesters on cleaning loss rate and impurity rate often tends to be contradictory. In this paper, an intelligent controlling model based on multi-objective optimization particle swarm (MOPSO) was constructed to solve this problem. The control model can real-time monitor the cleaning performance such as the cleaning loss rate and impurity rate and regulate the cleaning operation conditions such as the angle of the air distributor plate, the opening of the upper sieve and the fan speed. The field operation experiment of 10 kg feeding rice–wheat combine harvester proves that the control model based on MOPSO is more effective than the model based on fuzzy control. [ABSTRACT FROM AUTHOR]

To design a dialogue model and standard artificial programming interface (API), this paper designs an intelligent dialogue support for medical service systems and improving hospital intelligent serviceability, which combines patient pre-diagnosis, diagnosis, and post-diagnosis services with artificial intelligence depth. This is done via an intelligent man–machine dialogue support platform (MMDSP) suitable for medical services based on a multi-dimensional disease model and outpatient knowledge base with artificial intelligence. As a result of the intelligent service capability of the hospital service systems and platforms, patients' medical experiences have significantly improved. The platform standardizes the multi-channel service process, improves patient service efficiency, and reduces the cost of human resources and business knowledge learning. In addition, the integration of intelligent multi-system service information provides data support for hospitals to serve patients accurately and has good application and promotion value. [ABSTRACT FROM AUTHOR]

With the rapid development of social network platform, in order to purify the network environment, prevent the abuse of public opinion information, and control public opinion in a very short time, this paper proposes the application research of deep learning in social network public opinion sentiment recognition and analysis. Through the construction of social network model and social network knowledge map and the analysis of key technology, a network public opinion algorithm based on deep learning is proposed, and a competitive public opinion information communication model in online social networks is constructed, and then a simulation experiment is conducted on the improved model of the constructed social networks. The result shows that in public opinion management and control, for users with larger node degree, this technology can effectively understand and obtain public opinion information in a very short time with faster and wider information propagation speed, so as to realize the effective control of public opinion information and the processing of massive information. [ABSTRACT FROM AUTHOR]

The cleaning loss rate is a crucial performance metric for rice-wheat combine harvesters. Most studies on the relationship between the cleaning operation parameters of the combined harvester and the cleaning loss rate lack research on the dynamic correlation between harvest losses and operational parameters, specifically focusing on the dynamic correlation between harvest losses and multiple operational parameters such as the cleaning loss rate. In this paper, we formulate a dynamic deductive model for the scenario of harvesting loss regulation involving multiple operational parameters, delving into the dynamic associations between cleaning loss and the regulation of multiple operational parameters and constructing dynamic Bayesian network prediction model. Through experimental analysis of the dynamic Bayesian network prediction model for cleaning loss rates, the comparison of results among the other three algorithms demonstrated that it was efficient for our DBN to predict the cleaning loss rate. [ABSTRACT FROM AUTHOR]

In the medical field, analyzing various bone structures is crucial due to the rigid nature of bones. X-ray imaging plays an essential role in medical procedures, including bone age evaluation, fracture detection, and implant creation. However, operator involvement can introduce biases and increase processing time. Automating the process could reduce processing time and enhance diagnostic accuracy by minimizing biases and operator involvement. This paper introduces the Refined DeepLab model, a lightweight encoder–decoder-based approach for multiclass segmentation of hand bones. The primary objective is to assist physicians in tasks such as bone age analysis, fracture detection, hand movement analysis, and implant design. The research objectives are organized into three phases, with this work focusing on the first phase of our objectives, which is delineating bones from tissues, studying the bone structure, and multiclass segmentation of hand bones. The model utilizes DenseNet121 as its feature extractor and Sigmoid-weighted Linear Unit (SiLU) as its activation function. Experimental findings demonstrate promising performance in hand bone multiclass segmentation, with a Mean Intersection over Union (mIoU) of 85.02% and a Dice score of 92.2%. The comprehensive analysis of the results confirms that the proposed model excels, particularly in achieving the right balance between computational efficiency and precise segmentation. [ABSTRACT FROM AUTHOR]

Precisely forecasting the levels of PM 2. 5 is crucial for environmental conservation and human health. Thus, it serves as an essential indicator of atmospheric purity. In this paper, a PM 2. 5 concentration prediction model based on random forest and SHAP is proposed using air pollutants and meteorological conditions as the characterizing factors. Initially, pertinent information is gathered and subsequently manipulated, educated, and forecasted through the application of the random forest technique. Then, SHAP is used to explain the degree of influence of each feature in the model and the prediction results. Results of the experiment demonstrate that the random forest-based PM 2. 5 concentration prediction model for the three cities surpass the comparison model in the RMSE, MAE, and R 2 indicators. Examining SHAP values, the essential elements influencing the PM 2. 5 concentration are pinpointed. [ABSTRACT FROM AUTHOR]

Modern face detection algorithms fail to provide optimal results when they have to deal with larger amounts of data per frame while processing higher quality videos. This paper tackles that problem and offers a solution to deploy commercially used state-of-the-art face detection algorithms to process only the regions of interest in a frame, and discard the rest to decrease the data to be processed. The model maintains the accuracy of the base algorithm while decreasing the processing time per frame, thereby increasing the overall efficiency. The selection of region of interest is dependent on the detection of facial window in the previous frame. Therefore, the choice of base algorithm plays an important role in determining the speed of the framework. The model achieves increased processing speeds of about 69–76% more than the standalone usage of the detection algorithms for analyzed frame rates. [ABSTRACT FROM AUTHOR]

Mathematical morphology and convolution operators are two different methods to extract the characteristics and structures of images. Over the past decades, Deep Convolutional Neural Networks (DCNN) have been proven to be more powerful than traditional image-processing approaches. In this paper, we propose a novel structure called Deep Hybrid Neural Network (DHNN) by taking advantage of the convolution and morphological neural layers. Its practical application to polyp detection in medical images is illustrated. For experimental completeness, we adopt nine polyp image datasets, including publicly available data and our own collected data. For performance comparisons, we select three backbone models. Experimental results show that our DHNN achieves the best performance in comparisons in terms of computational complexity and accurate performance. [ABSTRACT FROM AUTHOR]

This paper focuses on the intelligent recognition problem of radar detection targets. Aiming at the low accuracy and slow speed of millimeter-wave radar clustering point cloud information, a feature algorithm of millimeter-wave radar suitable for detecting targets is proposed. In the detection of targets by millimeter-wave radar, distance is the biggest factor affecting the number and degree of sparsity. A method that combines the feature information of the point cloud with the KD tree proximity search algorithm and the DBSCAN clustering algorithm is proposed, which can adapt to the problems of uneven target point cloud, small amount of data and slow clustering speed. The improved algorithm can use the KD tree to quickly find adjacent points and calculate the distance between adjacent points. The corresponding number of thresholds is set according to the distance where the target is located, and the radius of the target area reflected by the millimeter-wave radar plus the distance of the last threshold point is used as the neighborhood radius of the improved algorithm. Therefore, fast and adaptive parameter adjustment of the millimeter-wave radar can be realized. Simulation tests show that the improved clustering algorithm has better parameters. The accuracy of the improved algorithm is increased by 4.2%, and it also greatly improves the clustering speed. [ABSTRACT FROM AUTHOR]

To improve the recognition ability of clinical named entity recognition (CNER) in a limited number of Chinese electronic medical records, it provides meaningful support for clinical advanced knowledge extraction. In this paper, using CCKS2019 Chinese electronic medical record as an experimental data source, a fusion model enhanced by knowledge graph (KG) is proposed, and the model is applied to specific Chinese CNER tasks. This study consists of three main parts: single-mode model construction and comparison experiment, KG enhancement experiment, and model fusion experiment. The model has achieved good performance in CNER from the results. The accuracy rate, recall rate, and F1 value are 83.825%, 84.705%, and 84.263%, respectively, which is the global optimal, which proves the effectiveness of the model. This provides a good help for further research of medical information. [ABSTRACT FROM AUTHOR]

In this paper, the method of embedding piezoelectric ceramic sensors is used to test the damage of the materials and models of the urban comprehensive pipe gallery. The monitoring signal is processed by HHT method, and the frequency and energy changes of the piezoelectric signal before and after the explosion were analyzed, thus the damage characteristics of the urban comprehensive pipe gallery under the internal gas explosion are determined. The results show that in the gas explosion test inside the pipe gallery model, the waveform and peak value of the piezoelectric signal before and after the explosion did not change significantly, indicating that there was no obvious damage to the side wall of the integrated pipe gallery. However, after further time-frequency analysis of the piezoelectric signal by HHT, it is found that the dominant frequency of the piezoelectric signal after explosion has a downward trend, so it is judged that there is a small damage in the pipe gallery after explosion. [ABSTRACT FROM AUTHOR]

Aiming at problems such as tracking failure caused by illumination changes often encountered during unmanned aerial vehicle (UAV) tracking, a target tracking algorithm with illumination adaptive and future-aware correlation filters is proposed based on the background-aware correlation filters (BACF) algorithm, which realizes reliable UAV tracking tasks at night. First, the dark scene is recognized, and an efficient image enhancement module is used to enhance the brightness of low illumination images. Then a future-aware module is constructed to train the tracking model using the contextual information of the target in the next frame for better robustness. Finally, the model updating stage involves adaptive filter updating and adaptive learning rate updating to enhance target tracking precision. The results of the comparison experiments with the state-of-the-art algorithms on the UAVDark135, UAVDT, and DTB70 datasets show that the algorithm in this paper outperforms the state-of-the-art tracking methods and has better tracking performance under light changes and fast motion (FM) scenarios. The tracking speed on a single Central processing unit (CPU) reaches 49 FPS, which satisfies the real-time requirement of UAV tracking. [ABSTRACT FROM AUTHOR]

The recognition of steel microstructure images plays a crucial role in the metallographic analysis process. Although some progress has been made through the application of artificial intelligence algorithms, several challenges remain. First, existing algorithms exhibit weak nonlinear feature extraction capabilities and noticeable limitations. Second, they overlook the intrinsic noise and redundant interference present in microscopic images. To address these issues, this paper investigates the automatic recognition of metallographic tissues by leveraging residual structures in deep neural networks. An enhanced residual network model based on transfer learning is proposed, which utilizes the pre-trained weights from the ImageNet dataset to facilitate learning with small sample data. This network offers higher classification accuracy and higher F1 scores. In addition, a deep residual shrinkage network model based on an attention mechanism is proposed. This model incorporates an attention sub-network into the original residual module and employs a soft threshold function to eliminate redundant features, including noise. The proposed algorithms are evaluated against various convolutional neural networks using 20 types of metallographic test sets. The experimental results showed that both methods have high accuracy rates of 95% and 94.44%, respectively, and F1 scores of 0.9464 and 0.9419. While maintaining the complexity of the model, there has been a significant improvement in accuracy, and the models exhibit strong generalization capabilities. Our research contributes to enhancing production efficiency, strengthening quality control, and improving material performance through computer vision technology. [ABSTRACT FROM AUTHOR]

Handwritten documents possess immense significance in domains such as law, history, and administration. However, they are vulnerable to forgery, which can undermine their credibility and reliability. This paper aims to establish a dependable technique for identifying altered text in handwritten document images, even in scenarios with high levels of noise and blur. Our study investigates 10 distinct categories of handwritten text that have been altered through various forgery operations. The suggested approach employs the deep neural architectures VGG16 and Resnet50 as feature extractors. The architecture comprises three parts: Feature extraction using individual models, a feature fusion layer, and a classification layer. Initially, we optimize the training process and feature extraction using VGG16 and ResNet50. The feature vectors obtained from both models are then fused together in the feature fusion layer and input into the classification layer for the classification task. Experiments are conducted on a custom-created dataset as well as benchmark datasets including ICPR FDC, IMEI Forged Number, and Kundu to demonstrate that the proposed method is superior to existing approaches. [ABSTRACT FROM AUTHOR]

Head Pose Estimation (HPE) has a wide range of applications in computer vision, but still faces challenges: (1) Existing studies commonly use Euler angles or quaternions as pose labels, which may lead to discontinuity problems. (2) HPE does not effectively address regression via rotated matrices. (3) There is a low recognition rate in complex scenes, high computational requirements, etc. This paper presents an improved unconstrained HPE model to address these challenges. First, a rotation matrix form is introduced to solve the problem of unclear rotation labels. Second, a continuous 6D rotation matrix representation is used for efficient and robust direct regression. The RepVGG-A2 lightweight framework is used for feature extraction, and by adding a multi-level feature fusion module and a coordinate attention mechanism with residual connection, to improve the network's ability to perceive contextual information and pay attention to features. The model's accuracy was further improved by replacing the network activation function and improving the loss function. Experiments on the BIWI dataset 7:3 dividing the training and test sets show that the average absolute error of HPE for the proposed network model is 2.41. Trained on the dataset 300W_LP and tested on the AFLW2000 and BIWI datasets, the average absolute errors of HPE of the proposed network model are 4.34 and 3.93. The experimental results demonstrate that the improved network has better HPE performance. [ABSTRACT FROM AUTHOR]