Proceedings of the 2nd International Conference on Networks, Communications and Intelligent Computing (NCIC 2024)

Inhaltsverzeichnis

1. Intelligent Computing

   1. Frontmatter

   2. Aviation Lock-Wire Twisting Direction Detection Based on CCF-YOLOv7 Model

      Yuguang Duan, Shiwei Zhao, Xibo Cao

      Abstract

      Aiming at the problem of low efficiency and inconsistent standards in visual inspection of aviation lock-wire twisting direction by human in typical maintenance scenarios, an automatic detection model CCF-YOLOv7 is constructed. Using YOLOv7 as the basic model, the convolutional block attention mechanism CBAM is integrated into the SPPCSPC spatial pooling pyramid block; Embedding a CA coordinate attention mechanism at the last three E-ELAN blocks of neck network; Optimizing the bounding box regression loss function CIoU to Focal-EIoU Loss. In addition, we also create an aviation lock-wire twisting direction dataset. Comparison and ablation experiments are conducted on the CCF-YOLOv7 model, and the results show that the CCF-YOLOv7 achieves the highest accuracy of 83.33%. Compared with existing object detection models like YOLOv5s, CCF-YOLOv7 achieves a better detection result in the dataset.

   3. A Decision-Level Fusion Algorithm for Infrared and Visible Images Based on Image Quality Assessment

      YanLi Yang, ZhiMan Liu

      Abstract

      In order to improve the localization accuracy of object detection algorithm, a decision-level fusion strategy based on infrared image and visible image quality is proposed. An object detection framework based on yolo-v3 model is established to detect infrared images and visible images obtained in the same scene respectively. To address the issue where the center point of a detected target may deviate from its true position due to poor image quality or insufficiently distinct target features, the gradient amplitude mean and Laplacian Operator variance are used as indicators to evaluate the image quality, and the fusion weight is assigned. The simulation results show that the proposed algorithm can fuse infrared image and visible image target detection results to improve object positioning accuracy. Compared with the detection results of a single target detection model, the fusion results adopted are 45.57~46.19% higher than the infrared image detection results, and 23.91~24.78% higher than the visible light detection results.

   4. A Hybrid Encoder-Decoder Based CNN Model for Improving Obstacle Detection Accuracy in USVs

      MD Asif Hasan, Haiming Chen, Di Wang, Changzhou Hua

      Abstract

      Obstacle detection is important for keeping safety of Unmanned Surface Vehicles (USVs). Currently, obstacles are usually detected by semantic segmentation using state-of-the-art encoder-decoder based deep learning models, like CNN (Convolutional Neural Networks). However, current CNN models face challenges when applied in marine environments due to their dynamic nature. Many existing studies have used Fully Convolutional Networks for the task of image segmentation. However, in these networks, the decoder path lacks sufficient low-level features from the encoder path for the proper reconstruction of the image, because large marine dataset for USVs are needed to train complex neural networks. By considering these issues, a hybrid encoder-decoder based CNN model, called HybridNet, is proposed for improving semantic image segmentation for USVs. The proposed encoder network uses pretrained model weight for faster learning and extracts multiscale features as in many SegNet models which are memory efficient. Then, some changes have been made in the decoder layer, which adds Atrous Convolution to increase the receptive field introducing dilation rate. Besides, we have added residual block connection as in U-Net models to compensate losing spatial information in the encoder layer during down sampling operation. The proposed model is tested and cross validated on the MaSTr1325 maritime datasets. Results show that HybridNet achieves higher precision, recall rate and mean Intersection over Union (mIoU 0.98 compared with baseline models like SegNet (Segmentation Network) and U-Net.

   5. LLM4SDP: Large Language Models for Software Defect Prediction

      Ge Jin, Xinjie Zhang, Xingmeng Shi, Haonan Tong

      Abstract

      Background: Cross-project defect prediction (CPDP) makes use of historical defect datasets gathered from source projects to train a model, which is then applied to a target project. Large language models (LLMs) have achieved promising performance on many fields including software engineering, e.g., code generation. However, existing cross-project defect prediction (CPDP) methods mainly center on traditional machine learning or deep-learning approaches and fail to explore the potential of large language models (LLMs). It is still an open question that whether LLMs are helpful for CPDP. Objective: To tackle this issue, we put forward a novel approach named LLM4SDP. This approach makes use of large language models to achieve cross-project prediction capabilities. As far as we know, this is the first time that large language models have been applied to cross-project defect prediction (CPDP) in the field of software defect prediction. Method: We investigate large language models including Qwen2-7b, Llama3-8b-instruction, and CodeGemma-7B-Chat. The well-known fine-tuning method LoRA is employed to adjust LLMs to help them better adapt to downstream tasks. We further explored the effect of over-sampling methods on LLM performance owing to the natural class imbalance of software defect datasets. Results: Experiments were conducted on five defect datasets (EQ, JDT, LC, ML, PDE), performance was evaluated using metrics like the Matthews correlation coefficient (MCC). Results show: (1) the proposed method yields varying performance across datasets. On the JDT dataset, the Qwen2-7b model achieves an MCC of 0.40 with SMOTE, a notable improvement. However, SMOTE decreases performance on EQ and ML. (2) SMOTE improves recall, especially in imbalanced datasets, but may reduce precision. (3) compared to baseline methods, the proposed approach demonstrates significant improvements in some datasets. Conclusion: It can be concluded that: (1) applying data balancing techniques like SMOTE is beneficial in certain datasets, especially for improving recall in imbalanced datasets (e.g., JDT, LC); (2) the proposed method shows promising results in some scenarios compared with existing CPDP methods, particularly with the CodeGemma-7B-Chat.

   6. A Trajectory Optimization Method for High-Sensing Performance in Urban Air Mobility

      Zhonghao Luo, Chunyu Pan

      Abstract

      Sensing ground users is recognized as one of the key operations in urban air mobility (UAM) systems, ensuring high sensing performance is crucial for providing accurate data for various urban tasks. In this paper, we focus on maximizing the radar sensing rate for each user by jointly optimizing the UAM aircraft’s flight trajectory, speed, and acceleration. The challenge lies in addressing the complex non-convex nature of the problem, which includes non-convex constraints that are difficult to solve using standard optimization techniques. To overcome this, we utilize methods such as successive convex approximation and Taylor series expansion, which enable us to transform the non-convex constraints into convex ones, making the problem solvable with convex optimization solvers. Simulation results validate the effectiveness of the proposed approach, demonstrating that it not only enhances the sensing rate but also exhibits robust performance across various flight conditions.

   7. Concurrent Graph Data Sharing and Resource Optimization Model in Edge Computing

      Weirong Xiu, Md Gapar Md Johar, Mohammed Hazim Alkawaz, Chen Bian

      Abstract

      With the development of the Internet of Things and smart cities, edge computing has gradually become a key technology for real-time data processing. The traditional cloud computing model has obvious limitations in data transmission and processing delay. This paper proposes a concurrent graph data sharing and resource optimization model, which aims to improve the resource utilization and computing efficiency of edge computing nodes in the process of concurrent execution of multi-tasks. By analyzing the data access pattern of tasks, the Longest Common Subsequence (LCS) algorithm was used to optimize the data sharing strategy, and the data read sequence was dynamically adjusted to reduce data transmission and repeated calculation. The experimental results show that the proposed model is superior to the traditional method in terms of task execution time, data transmission volume and resource utilization. The task execution time is shortened by about 33.3%, the data transmission volume is reduced by about 46%, and the resource utilization rate is improved by about 20%. This research provides an efficient solution for real-time data processing in edge computing environments, which holds significant importance for fields such as the Internet of Things (IoT), industrial automation, and smart cities.

   8. Energy-Efficient Computing Offloading and Resource Allocation for Mobile Edge Computing Enabled Vehicular Networks

      Jihang Shi, Jiaxuan Liu, Zhongyu Wang, Yingping Cui, Yubo Li, Zheng Chang, Guanghua Gu, Xuehua Li

      Abstract

      As vehicular networking advances, increasing interactions between vehicles and cloudlets have heightened the demand for computing resources. Mobile edge computing (MEC) addresses this challenge by utilizing edge resources to meet growing computational needs. This study introduces a communication scenario within vehicular networks, featuring an MEC-enabled roadside unit in interaction with multiple vehicles. We introduce an innovative approach that concurrently optimizes the allocation of power and bandwidth, to minimize overall energy consumption. The complexity of the original problem, attributed to the interplay of numerous optimization variables, is effectively managed by directly applying the Lagrange multiplier method to address the optimization challenges. Simulations demonstrate the proposed method achieves a substantial reduction in energy consumption, significantly outperforming existing benchmarks and highlighting its effectiveness.

   9. Double-Layer UAVs Network Design for Real-Time Distributed Collaborative Tasks

      Dianqing Meng, Yijun Guo, Xiaoshijie Zhang, Zijing Chen

      Abstract

      With the development of 6G, the use of unmanned aerial vehicles (UAVs) to assist the Internet of Things (IoT), has attracted tremendous attention due to its broad applications in recent years. This paper proposes the double-layer UAV framework for real-time distributed collaborative tasks, which need to be processed separately and then aggregated for processing jointly to form a global conclusion. We aim to minimize the total time delay of executing the tasks in the IoT network, including the data collection delay, distributed processing delay, data aggregation delay, and centralized processing delay, by jointly optimizing the location deployment, task offloading, and computing resource allocation of double-layer UAVs.To overcome the challenge of non-convexity in the original problem, we adopt a method that integrates block coordinate descent, successive convex approximation, and the PSO algorithm by breaking it down into three interdependent sub-problems, which are solved in an alternating manner until convergence is achieved. The simulation results indicate that the proposed framework substantially minimizes the overall time delay in UAV-assisted IoT networks compared to the baseline approach.

   10. Sample Based Contrastive Learning for DeepFake Detection

       Chen Shao, Fan Zhang, Jinxiao Wang, Benkui Zhang

       Abstract

       With the rapid advancement of deepfake techniques, this technology has attracted significant public attention. Traditional deepfake detection methods primarily focus on distinguishing between real and fake categories, often neglecting the sample level feature differences that lead to these distinctions. As a result, the feature representation distance between each fake sample and its corresponding original real sample is difficult to expand. This prevents the model from capturing discriminative features, ultimately limiting its performance. To address these limitations, we propose a novel deepfake detection framework, Sample Based Contrastive Learning for DeepFake Detection (SCL). SCL leverages the unique characteristics of the deepfake detection task by employing sample based contrastive learning to maximize the feature representation distance between real and fake sample pairs, thereby enhancing classification performance. Specifically, leveraging the temporal correlation of video frames, different frames from the anchor’s video are treated as positive pairs, encouraging them to be drawn closer to the anchor. In contrast, based on the characteristics of the deepfake detection task, video frames of fake or real samples that correspond to the anchor are treated as negative pairs, pushing them further away from the anchor. Experimental results on multiple deepfake detection datasets demonstrate the superior performance of the proposed method. The detection performance on several deepfake detection datasets have demonstrated the performance of SCL.

   11. Conditional Diffusion Model with T-UNet for Sketch Face Synthesis

       Jiyu Zhang, Kangning Du, Huanyu Bian, Pengcheng Wang

       Abstract

       Sketch face synthesis aims to create detailed and realistic sketch images from optical photos. Recently, diffusion models have effectively addressed challenges like over-smoothing and mode collapse by simulating the distribution of multi-channel input data, but due to limitations in capturing low-frequency information, the generated images lack authenticity in textural details. To address the concerns raised in the appeal, we propose the Conditional Denoising Diffusion Probability Model (DDPM) with The T-UNet. To improve the utilization of low-frequency information, we propose a T-UNet module. This module maintains UNet’s ability to capture high-frequency details while integrating the Transformer’s ability to process low-frequency information. As a result, it enhances the overall quality of the synthesized sketch images. Experiments on the CUHK dataset demonstrate that our approach generates sketches of excellent quality, surpassing existing methods in visual performance.

   12. Siamese Kolmogorov-Arnold Networks for Building Damage Assessment from Remote Sensing Image

       Zhengyang Yan, Lin Cao, Ying Chang, Xuan Liu

       Abstract

       Assessing building damage through high-resolution remote sensing images is crucial for humanitarian relief. The commonly used Siamese Network is inadequate for fitting complex features in building damage assessment. To address the abovementioned issues, this paper proposes Siamese Kolmogorov-Arnold Networks (SiameseKAN). In order to better accommodate the complex high-level features of multiple damage levels, a KAN Block is introduced at the bottleneck layer of SiameseKAN near the encoder-decoder architecture. In addition, the model proposes a rectangular field loss and an object gap loss to focus on the damage level characteristics within the building. The experimental results indicate that this method surpasses other methods, resulting in an improvement of the total score on the xBD dataset by 4.8%.

   13. Design and Implementation of Intelligent Dynamic Threat Assessment System for Ultra-Long-Range Targets

       Hu Xiang, Yichao Cai, Hui Zhang, Hao Li, Long Xiang

       Abstract

       An intelligent dynamic threat assessment system for ultra-long range air raid targets is designed, which realizes a breakthrough from theoretical research to practical application. After proving the superiority of the comprehensive weight method, the analytic hierarchy process (AHP) method and entropy weight method are selected to obtain the comprehensive weight vector through the least square optimization model. The hardware of the system is configured according to the requirements of information processing center and combat command center to meet the needs of actual combat. The design of the system software considers the characteristics of modern air combat, such as large number of targets, full types, fast maneuvering, and many emergencies. The design of hardware and software fully pay attention to the security confidentiality. The system has strong real-time performance and meets the requirements of dynamic evaluation. The effectiveness of the system is verified by simulation.

   14. Privacy-Preserving Offloading for Mobile Edge Computing: A Deep Reinforcement Learning Approach

       Qizhuo Yang, Yanhua Sun, Jiali Li, Zhuwei Wang, Chao Fang, Yang Sun

       Abstract

       Mobile edge computing (MEC) is now being utilized to address the growing demand for edge devices and high-throughput, low-latency computational tasks. Users can offload tasks to MEC servers, significantly reducing latency and energy consumption. However, traditional single-access-point networks often struggle to meet the needs of a large number of users, and issues such as privacy leakage during the offloading process and information theft on edge servers are frequently overlooked. This paper proposes a multi-access-point offloading framework and a privacy-preserving quality of service (QoS) model. We construct the framework and model, assess privacy risks and protection levels, and integrate blockchain technology to enhance information security. By comprehensively considering user experience and privacy protection, we model the problem as a Markov decision process (MDP). Furthermore, a multi-agent proximal policy optimization (MAPPO) algorithm is proposed to achieve the optimal offloading solution. Simulation results validate the effectiveness of the proposed algorithm.

   15. Class-Incremental Learning Framework Based on Out-Of-Distribution Detection with Masking Autoencoder Supervised by Classifier

       Shijie Zheng, Ziyang Li, Weirong Xiu

       Abstract

       Class-incremental learning (CIL) based on out-of-distribution (OOD) detection is a technique aiming at alleviating the problem of catastrophic forgetting. It enables the model continuously learn the features of new classes to accomplish new classification tasks without losing the ability to classify old classes as new classification tasks continue to arrive. Most existing autoencoder (AE) for OOD detection focus on reconstructing more image details, which doesn’t overcome the interference of background noise and suffers from CIL performance decline. Therefore, we propose a novel CIL framework based on OOD detection. Firstly, we utilize the classifier to supervise the training of the AE, which enables the model to capture more underlying features of in-distribution (ID) data and reduce the interference of background noise. Secondly, we generate masking vector by Taylor expansion to impose constraint on the latent space of the AE, which extract important information from latent vector while removing redundant information. Finally, to ensure stable training of our model, we adjust the proportion of important information in the latent vector during training our improved AE. The OOD analysis experiment results demonstrate that AE performs better than existing methods in OOD detection on CIFAR-100 dataset. Meanwhile, the CIL experiment results on MNIST, CIFAR-10, CIFAR-100, and TinyImageNet datasets demonstrate that our CIL framework performs better than the latest method.

   16. Introduing GNN-Transformer with Geospatial and Temporal Attention to MJO Prediction

       Zhewen Xu, Renge Zhou, Changzheng Liu, Jieyun Hao

       Abstract

       Madden-Julian Oscillation (MJO) is an atmospheric oscillation phenomenon that has a huge impact on global weather. To obtain more accurate and effective MJO prediction results, we propose a temporal forecasting deep learning model that combines dynamic graph neural networks with transformers. The model is designed to identify anomalous nodes during different time periods, and feed the results into a graph neural network for encoding, while a transformer model is employed for decoding, resulting in predictive outcomes. Experimental results demonstrate that our model achieves a maximum COR effective prediction value of 39 days and an RMSE effective prediction value of 31 days, outperforming existing models and surpassing the results of traditional partial differential numerical forecasting models to some extent.

   17. Driver Fatigue Detection Based on Multiple Behavioral Characteristics and Adjacent State Information

       Changbiao Xu, Jiao Liu, Zhifei Ming, Wenhao Huang

       Abstract

       A emerging trend in addressing fatigue driving is real-time monitoring of driver's fatigue state. In this paper, a driver fatigue detection scheme based on multiple behavioral characteristics and adjacent state information is designed. Linear weighting is applied to behavioral characteristics of driver's eyes, mouth and head to obtain state parameter. Then, the state parameter is fused with adjacent state information to get state judgment indicator that represents wakefulness, mild fatigue, severe fatigue, etc. The state judgment indicator and two sets of dual thresholds are used to achieve driver state detection. The innovation of the scheme is reflected in two aspects: the fusion of state parameter and adjacent state information and the introduction of two sets of dual thresholds. Simulation results show that compared with the scheme based on state parameter and single threshold, the scheme designed in this paper improves the accuracy by about 10%.

   18. RIS Assisted UAV Trajectory Planning Based on Deep Reinforcement Learning

       Shuo Chen, Zhongda Liu, Jianwei Wang, Xuehua Li

       Abstract

       Unmanned aerial vehicle (UAV) has demonstrated significant potential for applications in the new era of mobile network communications. However, the limitations of the resources carried by the UAV pose significant challenges to load capacity and operational sustainability. This paper presents a trajectory planning algorithm for UAVs, enhanced by a reconfigurable intelligent surface (RIS) and based on deep reinforcement learning. Initially, a system model for the RIS-assisted UAV communication network is developed. Subsequently, the study explores an optimization problem aimed at maximizing energy efficiency, with the UAV trajectory, transmission power, and phase shift as decision variables. Then, we propose a parallel TD3-based UAV trajectory planning algorithm assisted by RIS with optimally controllable phase shift (PTD3-RO). Both UAV and RIS are regarded as independent agents, which continuously optimize their behavior strategies through observation and learning. Finally, simulation results indicate that the PTD3-RO algorithm enables effective collaboration between UAV and RIS, and significantly enhances the energy efficiency of RIS assisted UAV communication system.

   19. An Improved Adaptive Weight Dynamic Window Approach Based on Imitation Learning

       Yi Li, Daxue Liu

       Abstract

       The Dynamic Window Approach (DWA) algorithm initially computes traversable areas using a kinematics model and subsequently selects the optimal trajectory based on an evaluation function, reward (ξ). This mode has emerged as a prevalent solution for decision-making in autonomous driving. However, the traditional DWA method employs a fixed evaluation function across all scenarios, indicating that it uses the same decision-making logic regardless of any driving situation, which is obviously unreasonable. In this paper, we propose an adaptive-weight DWA method based on imitation learning. The proposed method encompasses an algorithmic workflow that includes candidate trajectory generation, feature extraction, a trained model, and evaluation of optimal trajectories. Additionally, the paper presents the AdaptiveWeightNet architecture, which dynamically generates evaluation function weights tailored to the current driving scenario through encoding and computation of the scenario. These weights, combined with the generated candidate trajectory features, enable adaptive adjustment of decision-making logic to select the optimal trajectory. Our proposed method has demonstrated promising results on a self-constructed dataset. When compared to the fixed-weight results obtained using the Maximum Entropy Inverse Reinforcement Learning method, our improved algorithm exhibits substantial improvements in scene adaptability, driving stability, and similarity to human driving behavior.

   20. Multi-level and Multi-granularity Fusion Network for Multimodal Named Entity Recognition

       Yu Guo, Xiaoxu Hu, Gangyan Zeng, Xiaoyue Wu, Jing’ao Chen

       Abstract

       The vast flow of information on social media drives the spread of users’ opinions, and posts combining text and images contain rich information. In this context, the multimodal named entity recognition (MNER) task aims to identify entities and categorize them in social media posts using images. However, the textual expressions in those posts are often informal and lack contextual representation, making it difficult to determine entity types. Besides, accompanying images may either contain no relevant objects or have a significant amount of distracting information. To this end, we propose a Multi-level and Multi-granularity Fusion network (MMF) for the MNER task. First, multi-granularity multimodal features are considered to enhance the input representation. Then, we build a Transformer-based fusion network to achieve multi-layer semantic interaction between text and visual features at different granularities. Finally, the text features are connected with the obtained visual guidance, and conditional random fields (CRF) are used for entity label classification. Extensive experiments on the Twitter2015 and Twitter2017 datasets demonstrate the effectiveness and superiority of our proposed method.

   21. A Multi-agent Based Jointly Offloading and Caching Algorithm in Industrial IoT

       Hao Li, Xiaohuan Li, Xun Wang

       Abstract

       The advancement of Artificial Intelligence in recent years has driven the rapid growth of the Industrial Internet of Things (IIoT), bringing a significant increase in complex computing tasks and presenting a substantial challenge to the resource-constrained end devices (EDs) in IIoT. Edge computing and caching strategies have emerged as effective solutions to address this issue by transferring demanding computational tasks and storing frequently accessed data at Edge Access Points (EAPs), ultimately reducing latency and enhancing overall system efficiency. However, due to dynamic environments in IIoT, offloading requests are time varying and stationary offloading and caching algorithms are inefficient. As a result, this paper proposes a multi-agent based jointly offloading and caching (MAJOC) algorithm. Firstly, EAPs in an IIoT are treated as multiple reinforcement learning agents. Then jointly offloading and caching decisions of EDs and EAPs are modeled as Markov decision processes. We formulate the utility function which considers both system costs and cache hit ratio. At last, a MADDPG algorithm is used to maximize the utility function and achieve the optimal offloading and caching scheme in a dynamic IIoT scenario. Results of the simulation demonstrate that the MAJOC algorithm has a remarkable effect on task latency, reducing it by 27.2% and augmenting utility by 22.4%, when compared to benchmark techniques.

   22. A Multi-swarm Asynchronous Particle Swarm Optimizer with Enhanced Exploration and Exploitation

       Yang Zhang, Pan Zhang, Zhen Wang

       Abstract

       Global optimization algorithms, including the particle swarm optimization algorithm, have been widely used in solving complex optimization problems. However, particle swarm optimization frequently encounters the issue of premature convergence to local optima, which can result in suboptimal solutions. To enhance optimization capability, a multi-swarm asynchronous particle swarm optimizer (MSAPSO) with enhanced exploration and exploitation is proposed. MSAPSO divides the population into multiple subswarms, each of which executes its own mechanisms across three distinct phases. Moreover, asynchronous search strategies, including GPSO mechanism, asymptotic optimal guidance mechanism, exploration mechanism, and a Lévy-like mechanism, are designed to guide the velocity and position updates of particles in each subswarm asynchronously. This increases the exploration of more diverse search spaces while maintaining convergence towards potential solutions, significantly improving optimization performance. Experimental results on a variety of benchmark datasets demonstrate high search accuracy and competitive convergence speed, validating the efficacy of the proposed method.

   23. SG Filter with Statistical Feature Extraction Aids Deep Learning for Short-Range Wireless Sensing

       Haokun Zhang, Hui Zhao, Peng Wang

       Abstract

       In recent years, short-range wireless sensing technology based on Channel State Information (CSI) has become a popular research area. The application of deep learning has significantly improved sensing accuracy, but most existing research focuses on the selection of network models, with less attention paid to the impact of preprocessing and manual feature extraction on the results. To address this, we propose a method that combines SG filter with statistical feature extraction, aimed at enhancing the model’s efficiency in extracting features from CSI. In the preprocessing stage, we use an Savitzky-Golay (SG) filter to denoise the CSI signal, preserving the main components of the signal while filtering out noise interference. In the manual feature extraction stage, 12 statistical features are extracted from the SG-filtered CSI amplitude as input to the network. Experimental results indicate that this method performs exceptionally well in terms of sensing accuracy, computational complexity, and versatility across different networks. On the NTU-Fi public dataset, the accuracies for activity recognition and gait recognition were 99.69% and 99.12%, respectively, outperforming other existing sensing methods.

   24. Performance Evaluation of Scheduling Algorithms Using CloudSim

       Salem Omar Sati, Jiangjiang Zhang, Shanshan Tu, Muhammad Waqas

       Abstract

       Cloud computing is a growing method for using online resources, with varying traffic levels. Scheduling algorithms are crucial due to factors like execution time and latency. This paper examines cloud environments, highlighting models like SaaS, PaaS, IaaS, and types such as public, private, and hybrid. It uses CloudSim to simulate components like data centers and VMs. The study analyzes scheduling algorithms: FCFS, SJF, and RR. Results indicate SJF excels in most areas but not in makespan, FCFS fares well in makespan but is costly, and RR is efficient in makespan but slow in total completion time.

   25. User Satisfaction and Trust on Recommendation Systems with LLMs-Generated Explanations

       Luong Vuong Nguyen

       Abstract

       The movie recommendation systems have significantly enhanced users’ ability to discover new content. However, the opacity of these systems often leaves users questioning the rationale behind the suggestions. This study explores applying large language models (LLMs) to generate comprehensible and relevant explanations for movie recommendations. By integrating an LLM, specifically GPT-4, into a standard recommendation system, we aim to bridge the gap between complex recommendation algorithms and user understanding. The study involves a user-centric evaluation where participants interact with the system, receive movie recommendations, and are provided with LLM-generated explanations. User feedback is collected to assess the clarity, relevance, and overall satisfaction with these explanations. The results indicate that LLM-generated explanations significantly enhance user satisfaction and trust in the recommendation system compared to traditional methods. These findings suggest that incorporating LLMs can improve movie recommendation systems’ transparency and user experience, offering a promising direction for future enhancements.

   26. LLM-KBQA: A Knowledge Base Question Answering Framework Based on Large Language Models

       Ziliang Li, Haoliang Cui, Wen Zhang, Maosen Wang, Shaozhang Niu

       Abstract

       This article introduces a framework for knowledge base question answering using a Large Language Model (LLM). The framework transforms natural language queries into structured forms, enhancing accuracy and efficiency. Advanced fine-tuning techniques refine the LLM’s NLP capabilities. The framework also includes a novel knowledge matching approach combining coarse and fine granularity, leveraging character and word vector similarities with semantic analysis. Experimental results show high accuracy and efficiency in retrieving answers, offering a robust solution for information retrieval.

   27. Edge Computing Resource Optimization for UAV Collaboration in Sensing, Computing and Communication Networks

       Wenyue Jia, Chunyu Pan, Yafei Wang

       Abstract

       In order to solve the dependence of the traditional integrated sensing, computing and communication(ISCC) network on the ground, and to address the problem of high power consumption of ISCC network in emergency response and intensive mission scenarios, a fusion system for collaborative sensing by unmanned aerial vehicle (UAV) and edge computing is proposed. The problem of minimizing the total energy consumption of the system is established under the strict constraints of sensing performance, delay and task offload rate. Meanwhile, the problem is solved with deep reinforcement learning(DRL) and federated learning(FL). Simulations show that the scheme proposed in this paper outperforms other comparative schemes in energy consumption.

   28. DOA Estimation Based on BNF Constraints and Low-Order Processing in Impulsive Noise

       Weidong Wang, Yahui Zhang, Yongqing Zhang, Xingwang Li, Hui Li, Zhiqiang Liu

       Abstract

       The purpose of this paper is to address the performance degradation when using acoustic vector sensor array (AVSA) in impulsive noise environment for estimating direction of arrival (DOA). The existing methods of DOA estimation generally assume that the noise in the signal is Gaussian white noise. However, owing to the complexity of the underwater environment, the noise may contain impulsive characteristics, which will invalidate the original DOA estimation methods. In order to realize stable DOA estimation using AVSA in impulsive noise, a novel algorithm for sparse iteration based on bounded nonlinear function (BNF) and low-order processing is proposed in this paper. Firstly, the BNF is applied to suppress outliers caused by impulsive noise. Then, to overcome the limitation of BNF, which can only suppress impulsive noise in the nonlinear region, a sparse iterative technique based on low-order processing is applied. Finally, the DOA estimation is obtained through spectral peak search. The results of simulation indicate that the proposed algorithm can provide superior suppression of impulsive noise and significantly improves the performance of DOA estimation compared to existing methods. And this superiority is further outstanding when the generalized signal-to-noise ratio (GSNR) and the snapshots are small.

   29. Routing Delay Reduction Strategies for LEO Distributed Computing

       Zijia Ma, Zhuohang Li, Jinzhe Ruan, Ye Yao, Jiaen Zhou

       Abstract

       With the rapid development of the communication and computer industries, computational capability has become an important metric for assessing Quality of Service (QoS). Distributed computing within computational architectures is widely applied in mobile communications, satellite communications, and the Internet of Things (IoT). Edge computing and cloud computing are typical paradigms of distributed computing. Although cloud computing has its inherent advantages, it requires continually increasing deployment costs to maintain computational effectiveness in the face of large computational requests and transmission delays. Based on this, we propose a new paradigm where initial processing is conducted on data, followed by synchronous computations in both the cloud and edge devices. This effectively addresses complex problems while ensuring timely and efficient handling of simpler issues. We aim to minimize delays within acceptable energy consumption limits, thereby enhancing communication timeliness.

   30. Evaluating the Performance of the Diffusion-Based Molecular Communication Networks

       Jiaxing Wang, Xiaohui Bo, Wanjun Li

       Abstract

       Molecular communication (MC) represents a promising approach that employs molecules for facilitating communication between nanomachines. The diffusion-based MC faces significant challenges in terms of communication range due to the attenuation of molecular signals. To address this issue, an intermediate nano-machine is utilized as a relay between the transmitting nanomachine and its intended receiving nano-machine. In this paper, we introduce the amplify-and-forward (AF) relaying method to enhance the communication distance and ensure reliable long-distance communication in diffusion-based MC systems. We examine the impact of relay position and reception radius on system performance, utilizing the minimum error probability (MEP) detection criterion for signal reception. The performance of the MC is significantly impacted by the reception process of the information molecules. Simulation results indicate that the relaying scheme can effectively deduce the bit error rate (BER) by up to 11 dB when a large number of molecules are released.

   31. Transformer-Based Video Super-Resolution Algorithm with Adaptive Alignment Strategy Selection Methods

       Lei Zhang, Yujie Li, Xiaoming Tao, Nan Zhao, Fang Cui, Hengjiang Wang

       Abstract

       Traditional video super-resolution (VSR) methods, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), often face challenges in maintaining temporal consistency and dealing with complex motion dynamics. This paper proposes an adaptive VSR method that combines an adaptive alignment mechanism with the Swin Transformer to improve video quality across various motion scenarios. By utilizing the self-attention capabilities of Transformers, the proposed method effectively captures both spatial and temporal dependencies. The key innovation of this approach lies in the adaptive alignment module, which dynamically selects the optimal alignment strategy—static, patch-based, or guided deformable attention (GDA) based on the level of motion between frames. This adaptability allows the model to flexibly handle different motion scenarios, significantly improving frame coherence and detail reconstruction. Experimental evaluations on the Vimeo-90 K dataset demonstrate that the proposed method outperforms traditional CNN/RNN frameworks and existing Transformer-based approaches in terms of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM).

   32. A Decoupled Data Processing Based Measurement Method by Using Low-Cost MEMS Inertial Navigation Systems

       Ruofei Chen, Pingan Yan, Meng Xu, Zongmin Zhao

       Abstract

       Measurement plays an essential role in both production and daily life. With ongoing advancements in Inertial Measurement Units (IMUs) based on Micro-Electro-Mechanical Systems (MEMS), combined with Strapdown Inertial Navigation algorithms, the development of measurement devices has the potential to overcome the spatial limitations of conventional measurement tools. The primary advantage of inertial measurement systems lies in their ability to estimate their own position during motion, thereby preserving trajectories and enabling the extraction of distance information from these paths. However, low-cost MEMS IMUs are prone to considerable measurement error, as they introduce substantial noise and accumulate error over time, leading to reduced measurement accuracy. To address this issue, a linear acceleration sequence processing method based on two-level segmentation and standard deviation analysis is proposed. This method retains linear acceleration data as sequential information and minimizes error through segmentation, interpolation, and statistical techniques. An experimental validation platform built with low-cost MEMS inertial devices has demonstrated the feasibility of MEMS IMU-based measurements, with the proposed method improving short-term inertial navigation measurement accuracy on this platform.

   33. Green Energy-Aware Hybrid Service Scheduling Strategy

       Xingyu Xiang, Jinhe Zhou

       Abstract

       The application of Computing Power Network (CPN) in wide-area environments offers users efficient and flexible computational power supply services. To meet users’ computing demands while minimizing brown energy consumption, this paper proposes a Green Energy-Aware Hybrid Service Scheduling (GEHS) strategy. A detailed model encompassing computation, communication, and energy consumption is constructed, incorporating a differentiated cost strategy based on green energy supply states. By considering task computational preferences and latency sensitivity, the model intelligently schedules computing resources in CPN, leveraging heterogeneous resources (such as CPU and GPU) to handle diverse types of computational tasks. The proposed GEHS algorithm establishes a balance between task Quality of Service (QoS) and brown energy consumption, with experimental simulations validating the algorithm’s convergence and effectiveness. This optimized scheduling strategy ultimately achieves an optimal balance between resource utilization and green energy usage efficiency.

   34. Attention Based Gated Recurrent Neural Networks for Software Defect Prediction with Edge Computing

       Pengquan Liao, Ning Li, Mingzhe Liu, Wei Li

       Abstract

       Software defect prediction researches aim to find out potential risks in software projects by identifying files and dependencies. Deep learning models have been presented in defect prediction research, resulting in considerable achievements. Meanwhile, the recent growth in edge computing eases the high hardware requirement of machine learning methods. However, a number of irrelevant features in defect prediction data bring challenge to the performance of defect prediction models, while the lack in effective fusion of features decreases prediction accuracy. In order to improve these existing problems in deep learning defect prediction methods, we propose a defect prediction model suiting for edge computing via an attention-based gated recurrent unit neural network, in which the Bayesian algorithm is introduced for optimization. The model extracts both classic features from repository and semantic features from abstract syntax trees, fusing these features for comprehensive information. Furthermore, the attention component in our model guides the network to dynamically focus on key features for the defect prediction mission, thus overcome the issue of irrelevant features. In addition, the model is available for edge computing environment through federated learning. The experiment outcomes indicate that our method has comparatively good performance on software defect prediction dataset.

   35. Byzantine Fault Tolerance Based on n-ary Tree Communication Topology

       Jingbin Shi, Ning Li, Mingzhe Liu, Wei Li

       Abstract

       Practical Byzantine Fault Tolerance (PBFT) algorithms are renowned for their exceptional liveness, security, and fault tolerance. They possess significant value, especially within the domain of blockchain, with particular relevance to consortium chains. However, the current PBFT algorithms suffer from structural imperfections and high communication overhead. To address these issues, we propose a credit grouping PBFT algorithm based on n-ary tree communication topology (TCG-PBFT). Firstly, we propose a novel credit evaluation method to better align with the credit grouping approach employed in this study. Secondly, we rearrange the nodes, categorizing them into three groups based on credit scores: the primary node group, the consensus node group, and the observer group. Additionally, we employ a consistent hashing algorithm for the random selection of the primary node, replacing the previous method of entirely random selection across the network. Lastly, we enhance the message dissemination method in the consensus process by integrating credit grouping with a tree-structured mechanism, thus utilizing a tree-structured approach for message transmission instead of the original single-point broadcasting. Experimental results have shown that, compared to existing methods, the consensus algorithm structure proposed in this paper significantly mitigates the decline in throughput and the increase in latency as the number of nodes increases.

   36. MIMO-Based Resource Allocation Algorithm Using Semantic Importance in 5G Intelligent Vehicular Networks

       Hongzhi Luan, Xiaoman Cao, Yi Li, Chuanjie Qian, Qiong Mei

       Abstract

       Integrating 5G with the Internet of Vehicles (IoV) leverages Multiple-Input Multiple-Output (MIMO) technology to boost communication efficiency. However, current IoV systems emphasize intelligent task processing, which requires understanding semantic information, not just data transmission. Traditional resource allocation methods often fall short in addressing this, limiting the potential of 5G-enabled IoV. To address this, we propose a semantic importance-based resource allocation method. Using image classification as a case study, we assess the impact of semantic features on task performance to enable dynamic compression, reducing data load while supporting MIMO transmission. Our method combines one-dimensional enumeration with a Deep Q-Network (DQN) algorithm for joint optimization of semantic compression and resource allocation, maximizing vehicle task success rates. Experimental results show our approach significantly improves task performance by \(32.6\%\) compared to semantic-based methods and by \(185.5\%\) over traditional methods.

   37. PSO-Based Adaptive Controller Load Balancing Approach for Hybrid Cluster Architectures

       Zewei Zhang, Jiabao Chen, Jinyi Chen, Zhaoming Hu, Fangqing Tan, Haofei Xie, Chao Fang

       Abstract

       In high-dynamic networks, nodes are affected by mobility and changes in load, leading to issues such as uneven traffic distribution and sudden traffic spikes. To address the challenges mentioned above, the advantages of software defined networking (SDN), particularly the separation between the control and forwarding planes, are leveraged to design a cluster control architecture based on a globally distributed and locally centralized (GDLC) approach. This approach offers improved scalability and flexibility in managing network resources across clusters. Considering the dynamic network environment, we propose a network model and objective function for controller load balancing, aiming to optimize load distribution and minimize network congestion. By analyzing the load distribution of each controller within the cluster, a particle swarm optimization (PSO)-based load-balancing algorithm is proposed, which controls the merging/splitting decisions and switch assignment decisions between controllers, thereby addressing the load balancing issues of the controller cluster. Experimental results show that the proposed algorithm achieves better load balancing and lower migration costs compared to existing methods.

   38. An Anti-jamming Respiration Detection Method Based on Wi-Fi CSI in Wards

       Kexin Liu, Weina Gao, Jianfeng Song

       Abstract

       In wards, the respiratory rate of bedridden patients can significantly reflect their pathological status. Wi-Fi signals, with their secure, ubiquitous, and low-cost characteristics, have become an effective means for respiratory monitoring. This work introduces a method for respiration detection using Wi-Fi CSI (channel state information), characterized by the ability to use deployed Wi-Fi devices as signal sources. We have introduced an innovative approach that effectively suppresses interfering noise, thereby achieving results with high accuracy. Simulations indicate that our proposed method can achieve an error of less than 0.5 breaths per minute in real hospital ward scenarios.

   39. A Methodology to Improve Typhoon Track Prediction Effectiveness Based on Deep Learning-Powered Trajectory Correction

       Zhi Tao, Ruifan Chen, Zekai Su, Jiamin Che, Xin Dong, Jun Shen

       Abstract

       Typhoon track prediction is extremely important for disaster prevention and mitigation in coastal cities. The traditional numerical weather forecast WRF model can provide relatively accurate short-term forecast, but it is difficult to deal with the nonlinearity and complexity of typhoon track. This research focuses on applying deep learning approaches to refine WRF typhoon path predictions, seeking to improve accuracy by addressing the limitations inherent in conventional models. By combining the physical basis of NWP model with the nonlinear fitting ability of deep learning algorithm, mainly focusing on the Long Short-Term Memory network and transformer model, the best prediction performance can be achieved. The results of model validation experiments show that the hybrid method of WRF model and deep learning model can significantly improve the typhoon path prediction, and the prediction model using transformer performs better.

   40. An Improved YOLOv8 Method for Detection of Urban Appearance Violations

       Songlin Wei, Weiquan Li

       Abstract

       The detection of multiple types of urban appearance violations still faces several challenges, which make it difficult for algorithms to effectively extract the corresponding features. To address the above problems, a custom dataset was created including four categories of objects to solve the problem of lacking dataset. The CBAM attention mechanism was applied for improving the model's capacity to extract features. A tiny target detection head was added to capture the characteristics of small targets and context information more effectively, improving the performance of model. The application of the loss function WIoUv3 was made, which can adaptively modify the weight coefficient and enhance the bounding box's regression performance and detection resilience. Experimental results show that compared with the YOLOv8n model, the Precision, Recall, mAP_0.5 and mAP_0.5–0.95 of the proposed method increase by 3.8, 2.1, 3.3 and 4.8%, respectively. Although the FPS decreases from 135.1 to 107.5, it still maintains a relatively high running speed. Comparative experiments validate the effectiveness of the proposed method in object detection within practical scenarios of UAV images.

   41. Optimizing Resource Allocation Challenges in Cloud-Edge Collaborative Systems: A Deep Q-Learning and Genetic Algorithm Approach

       De-Yu Meng, Ye Jin, Wen-Xiang Li, Xin-Yi Huang, Ling-Xiao Cui

       Abstract

       In this paper, we find that an innovative approach to cloud-edge collaborative task scheduling optimization is presented, integrating Deep Q-Learning (DQN) with Genetic Algorithms (GA). By utilizing DQN for real-time decision-making and GA for global optimization, this method effectively handles challenges including latency, load balancing, and resource utilization. In addition, we designed a multi-agent task scheduling environment using OpenAI's Gym framework to model cloud-edge systems with dynamic task migration, aiming to balance loads and reduce delays. Experiments reveal that the combined method outperforms traditional algorithms—including Greedy Algorithm, Genetic Algorithm, Particle Swarm Optimization, and Simulated Annealing—in enhancing latency, load balancing, and resource allocation. These results validate the model's robustness in optimizing task distributions and adapting to changing workloads, which means providing an effective solution for complex cloud-edge computing challenges.