Information Technology and Control

Verification of 3D Electrical Equipment Model Based on Cross-source Point Cloud Registration Using Deep Neural Netwo

Hai Yu — 2024-12-21

With the popularization of digital twin techniques in power substations, assessment and verification of electrical equipment 3D models in digital twins according to as-built LiDAR point clouds become essential for the quality assurance of the designed substation models. However, computing the shape and texture differences between a 3D model and its corresponding point cloud is challenging due to the difficulty in aligning cross-source equipment point clouds with local geometric shape variations. In this paper, we propose a 3D model verification method based on overlap-aware cross-source point cloud registration. The key of the method is an overlap attention-based point cloud registration network with grouped KPConv, attention mechanism, and overlap-weighted circle loss. It improves the registration accuracy against local geometric shape variations between 3D models and LiDAR point clouds. In addition, due to the lack of real-world point cloud samples of electrical equipment, a novel point cloud augmentation method is employed for generating synthetic point clouds for improving the sim-to-real generalization capability of the network. Based on the pose alignment of the 3D model and the corresponding point cloud, a facet-level computing method is proposed for model differentiation and colorization. Experimental results using real-world point clouds of power substation equipment validate the performance of the proposed method.

Flexible Job Shop Scheduling Optimization with Machine and AGV Integration Based on Improved NSGA-II

Yong Liu — 2024-12-21

Aiming at the problem of integrated scheduling of machines and AGVs in a flexible job shop, this paper constructs a scheduling model with the optimization objectives of minimizing the maximum completion time, minimizing the machine load, and minimizing the total energy consumption. This model is based on a comprehensive consideration of the payload time and no-load time of AGVs between the loading and unloading stations and the machining machines. An improved NSGA-II algorithm is proposed to address this problem. The algorithm adopts a three-level coding structure based on processes, machines, and AGVs, and employs differentiated cross-variation strategies for different levels to enhance its global search capability. A variable domain search algorithm is introduced to boost the local search capability by combining different neighborhood search methods within the three-level coding structure. Additionally, reverse individuals are introduced to improve the elite retention strategy, thereby increasing the diversity of the population. Ultimately, the case test results demonstrate that the improved NSGA-II algorithm exhibits superior performance in solving the flexible job shop scheduling problem involving AGVs, and the effect of the number of AGVs on the scheduling objectives conforms to the law of diminishing marginal utility.

Instability Hazard Effect of Mined-out Areas Near the Mining Site by Fusion InSAR and PSO-BP Rock Mechanical Parameter Inversion

Liwei Yuan — 2024-12-21

Exploring the impact characteristics of near the mining activities on goaf and clarifying the disaster effects of instability in the mined-out area are critical research endeavors essential for effectively managing major risk hazards inherent to underground mining operations. This study integrates SBAS-InSAR and PSO-BP methodologies for inversely analyzing rock mechanical parameters in a lead-zinc deposit and applies the inversion results through the FLAC3D simulation method to the mining site adjacent to the null zone to study destabilizing disaster effects in the mined-out area under the influence of mining disturbance. The simulation aims to analyze the evolution process of surrounding rock destruction and instability in empty areas, identify the primary causes of disaster effects, develop a risk assessment and judgment model, and prevent accidents from occurring. The results of the study show that the integration of SBAS-InSAR and PSO-BP techniques for inverting rock mechanical parameters has yielded favorable outcomes in analyzing the destabilizing effect of the gob area near the mining site, and more accurately, it obtained the displacement and stress characteristics of the roof and pillars in the goaf under the mining disturbance as the mining near the empty area progresses. The simulation results demonstrate that influenced by mining disturbance, the maximum principal stress of the ore column in the void area significantly increases, primarily appearing as compressive stress. The distribution of the plastic zone indicates notably that the process of plastic deformation of the ore column leading to damage is primarily due to maximum shear stress. Evidently, the primary reason for the destabilization of the ore column is the concentration of stress resulting from mining disturbance, leading to compression and shear damage.FLAC3D simulation analysis has conclusively determined that pressure shear damage to the ore column resulting from undermining disturbance is the main cause of airspace destabilization in mining. The research methodology and analysis results provide vital theoretical support for the prevention and control measures against destabilization disasters in empty zones near mining sites, holding significant theoretical and practical value.

Enhanced Feature Extraction with AL-YOLOv9s Lightweight Model: Application in Key Component Recognition Within Highly Integrated Device Environments

Yang Wang — 2024-12-21

In environments containing highly integrated devices, accurately monitoring the status of circuit breaker lockouts is essential for maintaining the stability of power systems. Traditional detection methods are often inadequate due to complex equipment configurations and severe operational challenges. This paper presents an enhanced detection model, the AL-YOLOv9s, which improves the efficiency and accuracy of detecting circuit breaker lockouts. The AL-YOLOv9s model is based on the advanced YOLOv9s algorithm and incorporates an enhanced efficient multi-scale attention module to boost feature extraction capabilities. It also integrates channel and spatial attention mechanisms to optimize the feature fusion process, thereby improving detection performance. Additionally, the model has been optimized to a size of 4.7M, making it suitable for lightweight field applications without compromising accuracy. Experimental results demonstrate that the AL-YOLOv9s model achieves high standards in accuracy and portability, thus offering an effective and practical solution for lockout detection.

A Study on 3D Human Pose Estimation with a Hybrid Algorithm of Spatio-temporal Semantic Graph Attention and Deep Learning

Shengqing Lin — 2024-12-21

This paper introduces a method to enhance 3D human pose estimation accuracy by leveraging human topological structure and temporal information, addressing inaccuracies due to occlusion and complex poses. It proposes a spatiotemporal Transformer network that aggregates local temporal information to predict 3D poses for video frames, reducing sequence length through cross-step convolution. To further handle occlusion and information loss, the paper suggests a spatiotemporal graph attention network that incorporates spatial constraints and graph convolution with an improved adjacency matrix to emphasize local information in pose inference. A temporal convolutional network is also employed to model time, and the network alternates between temporal and spatial attention modules to prevent spatiotemporal information loss. Experiments on Human3.6m and HumanEva datasets demonstrate that the proposed method outperforms other approaches in prediction accuracy.

Integrating Deep Learning into Educational Big Data Analytics for Enhanced Intelligent Learning Platforms

Min Zhang — 2024-12-21

Exploring the field of educational big data analytics and gaining insights into student behaviour and its connection to academic performance is crucial for creating intelligent learning environments. Technological innovations have changed how students learn and reshaped the nature of education. Technological advancements have unquestionably made learning more accessible, faster, and enjoyable for pupils. When deep learning is integrated with learning management systems, intelligent course content may be generated with high accuracy, and no human interaction is required. This study utilises advanced deep learning techniques to analyse the xAPI-Educational Mining Dataset and reveal valuable insights that can significantly improve online learning experiences. The study underscores the crucial importance of parental involvement, emphasising its link to student attendance and overall satisfaction with the educational institution. In addition, the results suggest that students who actively participate in course announcements and utilise resources tend to achieve better academic outcomes, highlighting the significance of resource utilisation in achieving academic success. On the other hand, engaging in conversations seems to have a minimal effect on how students are categorised. Building upon these findings, a novel predictive model is introduced, utilising Long Short-Term Memory (LSTM) networks. This model utilises sequential student interaction data to predict future behaviour and academic outcomes, helping online platforms understand student actions and make informed decisions. This study makes a valuable contribution to developing cutting-edge intelligent learning approaches. It achieves this by utilising the potential of educational big data analytics and deep learning techniques.

Imaging Segmentation of Brain Tumors Based on the Modified U-net Method

Yajie Zhang — 2024-12-21

Brain tumor segmentation in medical image analysis is a challenging task. Deep learning techniques have recently shown promise in resolving a variety of computer vision problems, such as semantic segmentation and image classification. Brain MRI (magnetic resonance imaging) requires precise brain image segmentation for effective, rapid diagnosis and treatment planning. However, it is quite difficult to manually segment the brain image rapidly and accurately in low-quality, noisy images. This paper proposes a U-Net and combined attention mechanism-based method. This research enhances the segmentation of images of tumors in the brain using modified U-net. Traditional U-net segmentation techniques are still widely used in the medical field, but they have a number of limitations when dealing with small targets and fuzzier boundaries. To address this issue, we made the following modifications to U-net: We propose attention mechanisms to assist the network in concentrating on important regions. The multiscale feature fusion strategy improves the efficacy of network segmentation at various scales. Cross-entropy loss function and data augmentation improve the performance of the network further. Our method was validated using the Brats2019 dataset. The experimental results demonstrate that our proposed methodology exhibits superior speed and efficiency compared to existing techniques in the context of brain image segmentation. The dice coefficients for the multiple branch TS-U-net model were 0.876, 0.868, and 0.814 in the tumor subregions of WT, TC, and ET, respectively. This exemplifies the feasibility and potential of our methodology for the segmentation of medical images.

The Robust Asymmetric Minimum Cost Consensus Models with Interval-type Opinions

Zexing Dai — 2024-12-21

Existing consensus models primarily rely on precise opinions from decision-makers in a predefined context, neglecting the dynamics of expert opinion adjustments. To address this limitation, we introduce the interval-type opinions and explores the group decision consensus model under asymmetric adjustment cost from an uncertainty perspective. Then, the robust optimization theory is applied to adress the uncertainty in adjustment costs of decision-making individuals. The robust asymmetric cost consensus model of interval opinions under three uncertain scenarios is built. Finally, the validity of proposed model is verified by numerical calculations, and a sensitivity analysis and comparative study are performed. The results show that: (1) Utilizing interval opinions can significantly reduce consensus costs when compared to precise opinions; (2) Comprehensively comparing the three proposed robust models, the consensus model with budget asymmetric cost has the best performance.

Contour Detection by a Dark-Adaptation Model

Wei Zhou — 2024-12-21

The color contour detection model used for simulating the cone photoreceptor cell- lateral geniculate nucleus (LGN) – primary visual cortex (V1) visual pathway has achieved reliable results. In contrast, the rod photoreceptor cells employ a dark adaptive mechanism, which plays a key role in contour extraction in poorly lit environments. We employ this mechanism to propose a bionic model for contour detection. The proposed model divides the dark adaptation process into several stages and extracts the image information at each stage for subsequent integration. For evaluation, we applied the proposed dark adaptation model as the front-end processing method of the gray and color contour detection model, and performed experimental verification on the RuG, BSDS300/500, and NYUD databases. In comparison with a similar state-of-the-art model, the detection performance of the proposed model has several advantages; in particular, it extracts contour information more effectively in interior scenes lit with dim colors.

Point Cloud Upsampling Network Incorporating Dynamic Graph Convolution and Multi-Head Attention

Xiaoping Yang — 2024-12-21

To address the problems that graph convolution uses a fixed graph structure, fails to capture dynamic or changing graph structure information, and is prone to bias by employing the same attention. A point-cloud upsampling network (DGCMSA-PU) incorporating Dynamic Graph Convolutional (DGCNN) and Multi-head Self-Attention (MHSA) is designed. DGCNN is utilised for up-sampling and a MHSA mechanism is incorporated to simultaneously fuse information from different attention heads. The edge relationships between nodes in the graph data are captured by edge convolution (EdgeConv), and the graph structure is dynamically constructed based on the relationships between nodes. Then the features of the point cloud are extracted by the three attention heads with different weights and different foci. Finally, an up-down-up structure is used to extend the features and reconstruct the 3D coordinates of the output point cloud. The superiority of DGCMSA-PU in the up-sampling task is verified through experiments comparing it with existing up-sampling networks, and the robustness of the network to noise and varying number of input point clouds, as well as the important role of the Multi Headed Attention module in the performance improvement of the network, are analysed through robustness and ablation experiments.

Robust Incentive Mechanism of Federated Learning for Data Quality Uncertainty

Chao Wang — 2024-12-21

In order to make the incentive mechanism more suitable for the actual training situation and improve the eﬃciency of the model, the robust incentive mechanism of federated leaning is proposed to deal with uncertainty of the data quality. (1) Firstly, the incentive mechanism of federated learning is constructed by the use of Stackelberg game to optimize the central server and data owner utilities, respectively. (2) Secondly, the uncertainty of data quality of the data owners is present by two robust uncertainty sets, and the corresponding incentive mechanism of the robust Stackelberg game is given. (3) Thirdly, the existence of equilibrium solution of the game is proved and the equilibrium solution of the whole game is derived. (4) Finally, the feasibility and robustness of the model are veriﬁed, and in the comparative experiments, the central server can select the optimal combination of perturbation ratio and uncertainty level according to the preference for uncertainty risk to obtain the optimal incentive mechanism. The incentive mechanism designed in this article not only considers the uncertainty in actual training, but also has a good incentive eﬀect on model training under diﬀerent risk preferences.

Smartphone-Based Psychological Sensing: A Large-Scale Study on the Impact of Extreme Isolation

Zhaohui Yuan — 2024-12-21

The COVID-19 pandemic and associated isolation measures have greatly impacted mental health, especially among students. Previous attempts at using mobile sensors to analyze users' emotional states faced barriers including insufficient data and limited modalities. This study aims to address these limitations and derive insights on psychological changes under extreme isolation. We collected a large-scale multivariate dataset from 725 undergraduate students during the complete COVID-19 campus lockdown period. To our knowledge, this is the largest dataset on this population during an extended isolated period. Features were engineered from mobile sensor data to capture modalities including physical activity, sleep patterns, and social interaction. Additionally, self-reported assessments related to mental health conditions were compiled. This rich dataset was leveraged to develop a machine learning model based on autoencoders to detect emotional states. Comprehensive experiments indicate the model can accurately predict mental health changes using mobile sensor data. Our work has unique contributions in collecting large-scale isolated data, engineering informative modalities for modeling mental health, and providing a validated detection method. This can support rapid screening and intervention for mental health crises, especially those emerging from extreme events. The dataset and models open promising directions for big data analytics in mobile health and psychological research.

Green Rural Modern Architectural Design Based on Pose Recognition Algorithm of Thermal Discomfort

Xiaomei Gao — 2024-12-21

To reduce the energy waste of modern rural buildings caused by over-cold or over-heat supply, this paper presents a method to realize energy-saving design of modern green rural buildings by using thermal decomposition location recognition algorithm. Based on the key points of the human skeleton, the pose recognition framework is constructed, and the deep learning network is combined to detect the human thermal disturbance posture. Furthermore, an end-to-end thermal inauthentic pose recognition algorithm is proposed to establish a green intelligent building energy minimization model considering thermal comfort range. The results show that the recognition rate of 1D convolution +LSTM model is 100%, and the optimal accuracy of 16 frames of image sequence with decoder module is 92.052%. Compared to traditional algorithms, the method can save up to 10% of the total energy cost and reduce the total temperature deviation. This study is of great significance for intelligent control of indoor thermal environment and improvement of energy utilization efficiency.

MS-UNet: A Novel Multi-scale U-shaped Network for COVID-19 CT Image Segmentation

Shangwang Liu — 2024-12-21

The U-Net network has its own powerful capabilities in medical image segmentation tasks, yet still it is a challenging task to make U-Net accurately segment the infected lesions of COVID-19 CT images because these lesion areas are usually irregular in shape, various in size, and blurry in boundaries. In this paper, a novel multiscale U-shaped network based on U-Net for accurate segmentation of lesion regions in COVID-19 CT images is proposed. First, we generate two auxiliary scale features (f_i^0.5, f_i^1.5) based on the main scale feature (f_i^1.0) through zoom strategy. Secondly, we design the Scale Integration Module (SIM), which is capable of filtering and aggregating scale-specific features and can fully exploit multi-scale semantic information. Thirdly, the hierarchical mixed module (HMM) has successfully substituted for the down-up aggregation process of the U-Net network, which further enhances the mixed scale features. On the dataset COVID-19-CT829, compared with the recent COVID-19 segmentation model, hiformer, the Dice, Sen and F-measure of our network have increased by 2.24%, 2.83%, 3.14%, respectively; on the dataset COVID-19-CT100, the Dice, Sen and F-measure of our network have increased by 2.91%, 3.72%, 2.42%, respectively. Moreover, we have validated the generalizability and portability of our network on other medical datasets (Polyp segmentation dataset: CVC-612 and kvasir), and our network has also achieved superior results of COVID-19 CT image segmentation.

PMF-YOLOv8: Enhanced Ship Detection Model in Remote Sensing Images

Dan Chen — 2024-12-21

Satellite remote sensing technology plays a pivotal role in ship monitoring at sea, with ship detection by artificial intelligence becoming the primary means. However, due to the intricate marine environment and the similarity between classes of remote sensing ships, the detection of remote sensing ships still faces significant challenges. Existing detection models tend to overlook the loss of fine-grained features of remote sensing ships during the deepening of the network. To address this issue, we proposed an enhanced Pyramid for Multi-Scale Feature Fusion (PMF) to optimize the YOLOv8 algorithm. After incorporating a fusion of shallow-level features into the neck portion of YOLOv8, an adaptive spatial feature fusion approach coupled with a path aggregation network was employed to process the output features of the backbone network. This integration enhances the learning of fine-grained features and addresses the issue of feature loss, a common challenge in existing networks. Furthermore, to enhance feature extraction, we introduced an enhanced R-C2f module. Finally, Inner-MPDIoU was employed as the bounding box loss to address the issue of missed detections that may arise in the context of dense remote sensing ships. Experiments were conducted on FGSC-T, a dataset comprising 22 classes of ships, to assess the efficacy and viability of the algorithm. In comparison to the original YOLOv8, the mAP50, mAP50-95, Recall, and Precision increased by 3.7%, 4.1%, 5.7%, and 2.5%, respectively. Furthermore, the detection speed of PMF-YOLOv8 can reach 74 fps, which meets the requirements for real-time detection of remote sensing ships.

Deep Learning-Based Trajectory Tracking Method for Intelligently Network-Connected Driverless Vehicles in Narrow Areas

YaJun Han — 2024-12-21

Driverless vehicles are the development direction of intelligent transportation. In recent years, the rapid development of driverless transportation technology, especially the practical performance of intelligently network-connected driverless vehicles has improved rapidly. However, due to problems with traffic planning, many roads are still relatively narrow. When an intelligently networked driverless car moves in a narrow area, the lack of precision in trajectory tracking can easily cause traffic accidents due to small trajectory changes. In this paper, for the driving characteristics of intelligently networked driverless vehicles in narrow areas, an improved Faster R-CNN target detection network is proposed that introduces a deep residual network ResNet-50, a dual attention mechanism CBAM, and an ROI-Pooling to estimate the position information of driverless vehicles in the video of the traffic scene. Based on the target detection results of driverless vehicles and the appearance characteristics of vehicles, the novel DeepSORT vehicle tracking algorithm improved by OSNet full-scale network and complete intersection over union (CIoU), is employed to derive a vehicle trajectory within a single camera on a real road. The UA-DETRAC dataset in real scenarios is selected to run experiments, and the results demonstrate that the proposed target detection and tracking algorithms perform well, and effectively realize target detection and trajectory tracking of intelligently internet-connected driverless vehicles in narrow areas, which can help realize the further performance enhancement. The improved DeepSORT achieves an impressive MOTA of 96.1% and MOTP of 0.115.

Internet Finance Non-stationary Time Series Prediction Algorithm Based on Deep Learning

Yangyi Li — 2024-12-21

Inaccurate prediction results of financial time series will lead to wrong investment decisions. Therefore, a prediction algorithm for Internet financial non-stationary time series based on deep learning is proposed. EMD (empirical mode decomposition) method is used to divide the collected historical Internet financial non-stationary time series information into high-frequency and low-frequency parts, and remove the noise in the decomposed high-frequency components to obtain the financial non-stationary time series without noise. The knowledge map method is used to mine the transaction characteristics and market characteristics of Internet finance from the financial non-stationary time series without noise, and the two are fused as the input of the improved CNN (convolutional neural network) prediction model. The prediction results of Internet financial time series are obtained through CNN. The experimental results show that after setting the CNN parameters, the predicted results are consistent with the actual market trends. The highest RSE of the predicted result is 0.551, The highest RAE is 0.443, which is relatively low, the CORR value is 0.864, which is relatively high, indicating that the relative square root error, relative absolute error, and relevant empirical coefficients of the prediction results are all good, making it a highly applicable algorithm.

A Survey on Privacy Attacks and Defenses in Graph Neural Networks

Lanhua Luo — 2024-12-21

Graph neural networks (GNNs) have emerged as a powerful tool in the field of graph machine learning, demonstrating by a various practical applications. However, the complex nature of graph structures and their expanding use across different scenarios present challenges for GNNs in terms of privacy protection. While there have been studies dedicated to addressing the privacy leakage problem of GNNs, many issues remain unresolved. This survey aims to provide a comprehensive understanding of the scientific challenges in the field of privacy-preserving GNNs. The survey begins with a succinct review of recent research on graph data privacy, followed by an analysis of the current methods for GNNs privacy attacks. Subsequently, the survey categorizes and explores the limitations, evaluation standards, and privacy defense technologies for GNNs, with a focus on data anonymization, differential privacy, graph-based federated learning, and methods based on adversarial learning. Additionally, the survey also summarizes some widely used datasets in GNNs privacy attacks and defenses. Finally, we identify several open challenges and possible directions for future research.

Image Denoising Using Adaptive Weighted Low-Rank Matrix Recovery

Yujuan Wang — 2024-12-21

This paper introduces a new image denoising method using adaptive weighted low-rank matrix recovery to tackle the challenges of separating low-rank information from noise and improving performance affected by empirical hyperparameters. We start by using image nonlocal similarity to build a low-rank denoising model, then apply the Gerschgorin theory to precisely determine the rank of the low -rank matrix. With this rank estimation, we use adaptive weighting along with singular value decomposition and weighted soft-thresholding to solve the denoising model, resulting in the denoised image. Experiments show our algorithm surpasses traditional denoising methods in average PSNR and SSIM. Specifically, for images contaminated with high-intensity noise (with a variance of 100), our algorithm achieves average PSNR and SSIM values of 24.66dB and 0.7267, respectively. Additionally, our algorithm exhibits superior performance in denoising images with real noise and is also applicable to color image denoising.

Fish Catch Sorting and Detection Model Improved Based on YOLOv8 Model

Ping Yang — 2024-12-21

The primary challenge in trawl fishing lies in its limited selectivity, resulting in highly diverse fish catches with severe mixed-species issues. The catches from trawl fishing require manual sorting, leading to low work efficiency and high labor demands. In order to tackle this issue, the present study introduces an enhanced version of the DWR module by utilizing DilatedReparamBlock convolution, introducing a novel dilated convolution module. This module is integrated into the YOLOv8 model, enhancing its capacity to capture features from the enlarged receptive field in the upper network layers. Furthermore, a new attention mechanism based on a multi-branch structure with the CA attention mechanism is incorporated into the YOLOv8 model. This attention mechanism fully extracts image features, enhances feature representation, strengthens the generation of offsets and sampling weights, and improves the accuracy of target recognition. Lightweight improvements to the detection head are achieved through the use of shared convolutions, ultimately resulting in a significant reduction in the number of model parameters. Our empirical findings indicate that the refined model exhibits a 2.2% enhancement in mAP@0.5 when benchmarked against the initial YOLOv8 model, Offering a significant reference for the advancement of an effective embedded system for fish sorting.