Journal of Cyber Security and Mobility

Splicing Tampering Detection Algorithm Design for Digital Media Image Privacy

Yuxuan Liu — 2026-08-09

To make image information more authentic and complete and to prevent splicing tampered image data from affecting social fairness, this paper proposes an algorithm based on Deep Convolutional Neural Networks to detect splicing tampered digital media image privacy. Building upon deep convolutional feature extraction, this algorithm introduces a self-attention mechanism to enhance focus on tampered regions. For the first time, it innovatively applies a boundary-aware loss function to patch tampering detection, effectively addressing the issue of ambiguous boundary region detection and significantly improving localization accuracy. The experiment was conducted on datasets from the Institute of Automation, Chinese Academy of Sciences, Cover Dataset, National Institute of Standards and Technology datasets, and the 2020 Image Tampering Dataset. The algorithm demonstrated the following performance metrics: area under the receiver operating characteristic curve values of 0.971, 0.961, and 0.987, respectively; accuracy rate of 98.94%; precision rate of 97.12%; recall rate of 99.16%; F1 mean values of 0.941 and 0.952 under gamma ray and noise interference, respectively, indicating strong robustness. These results prove that the proposed algorithm can achieve precise detection of splicing tampered image privacy. It effectively addresses the problem of insufficient detection accuracy in some existing methods. It also promotes the intelligent development of detection and contributes to building a more authentic information environment in society.

Social Network Privacy Protection Based on Differential Privacy Technology and Community Discovery Algorithm

Xia Wu — 2026-08-09

The high aggregation of user relationship and behavioral data in social networks continues to aggravate privacy leaks. How to strike a balance between privacy protection and data availability has become a research hotspot. To collaboratively optimize user information security and community structure identification, this study proposes a social network privacy protection model that integrates differential privacy technology and community discovery algorithms. First, a differential privacy noise injection mechanism is constructed to perturb node data and combine it with blockchain storage to ensure that the data cannot be tampered with. Then, a community division strategy based on information entropy and mutual information is introduced to achieve high-precision community identification through modularity optimization. The accuracy of the proposed model reached 98.1% when the data set size was 800, which was about 3.4% and 9% higher than that of other models, respectively. The root mean square error was 8.2, which was about 20% lower than that of the traditional model. The convergence speed was increased to 380 iterations, which was about 15% faster than that of the comparison algorithm. The privacy protection strength and scalability scores reached 9.3 and 9.5, respectively. The simulation test results showed that, under different data types, the accuracy of the model grew from 0.87 to 0.98, and the F1 value grew from 0.84 to 0.95. The integration of differential privacy and community discovery effectively improves the privacy protection strength and structural analysis accuracy of social networks, providing a highly feasible solution for multi-scenario social data security analysis.

A Comparative Study of Deep Learning Models for Malware Detection in IoT Networks: CNN, LSTM, and Hybrid Architectures

Yuan Liu — 2026-08-09

The rapid expansion of the Internet of Things (IoT) has intensified security concerns. Many IoT devices operate with limited computational resources and inadequate built-in protection. This makes them vulnerable to malware attacks such as Distributed Denial-of-Service (DDoS), botnets, and ransomware. Traditional signature-based detection techniques struggle to identify evolving and previously unseen threats, highlighting the need for intelligent detection mechanisms. This study proposes a Hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) model for effective IoT malware detection. The CNN component extracts spatial features from network traffic, including packet-level and flow-based characteristics. The LSTM component captures temporal dependencies and sequential attack patterns. By integrating spatial and temporal learning, the hybrid architecture enhances detection capability for both known and zero-day attacks. Experimental results demonstrate superior performance compared to standalone CNN and LSTM models, achieving 99.92% accuracy, precision, recall, and F1-score, with a ROC-AUC of 0.999703.

Long-Distance Cross-Domain English Online Resources Based on Attribute-Based Encryption

Xiaofeng You — 2026-08-09

The rapid growth of distributed learning systems and cross-disciplinary digital services development has augmented the pressure on protecting and effective systems of sharing online educational resources. Conventional identity-based access control systems usually do not offer scalable and fine-grained security in heterogeneous systems where users are members of various administrative domains. To overcome these weaknesses, this paper suggests a secure long-distance cross-domain resource sharing model that is founded on a Hybrid Advanced Encryption Standard (AES) and Multi-Authority Key-Policy Attribute-Based Encryption (MA-KP-ABE) scheme of English online learning resources. The offered architecture combines symmetric encryption as the efficient method of data protection with the attribute-based encryption to implement flexible access control policies across various domains. Within the framework, AES is employed in expedited data encryption, MA-KP-ABE is employed in efficient key management, attribute verification by multiple authorities, dynamic policy enforcement as well as the ability to revoke attributes. The system was tested experimentally on some performance measures such as encryption time, decryption time, key generation time, storage overhead, scalability and accuracy of access control. The experimental findings prove that the proposed hybrid framework has an encryption time of 105.2 ms, decryption time of 82.5 ms, and a key generation time of 52.59 ms. The model also has low storage overhead of 6.5% and scalability to cross-domain settings of over 1500 users. Moreover, the system has a high accuracy in access control of 98.60% which means strong authorization and policy enforcement. The innovation of the given work is formed by the combination of AES and multi-authority KP-ABE with dynamic policy management and attributes revocation where it is possible to provide cross-domain access to online educational resources with the required security, scalability, and efficiency.

Federated Learning Privacy Protection Collaboration Model Based on Improved Cyber Threat Collaborative Analysis Capability

Rui Zhuang — 2026-08-09

Traditional federated learning has challenges such as slow model convergence, lagging situational awareness, and risk of gradient privacy leakage in cyber threat collaborative analysis capability. This study proposes a federated learning privacy-preserving collaboration model based on threat intelligence drive and hierarchical aggregation. The model optimizes participating nodes through a dynamic client selection mechanism, uses a hierarchical aggregation strategy to balance the learning of basic features and advanced threat patterns, and introduces an adaptive differential privacy mechanism to strengthen gradient protection. The experiment is based on the CIC-IDS-2018 public dataset, which covers various types of attacks in real network environments, with a data volume of approximately 5 million pieces. It is divided into 50 clients in a non-independent and identically distributed manner to simulate cross organizational collaboration scenarios. In the threat detection task, the model designed by the research institute achieved an accuracy of 91.8% and an F1 Score of 89.8% compared to baseline models such as FedAvg, FedProx, and DP FedAvg. All indicators were superior to the comparison model. In addition, in the advanced persistent threat attack scenario, the attack chain detection rate increased to 95.6%, and the average detection time was shortened to 2.8 hours. In terms of privacy protection, the Rényi privacy loss was only 2.89 with a budget of ε=3.0. The proposed model effectively improves the efficiency and timeliness of collaborative detection of cross-organizational threats while ensuring data privacy and provides a feasible solution for building a safe and reliable collaborative defense system. It should be pointed out that while improving detection performance, the model introduces additional communication overhead caused by hierarchical aggregation and dynamic selection mechanisms. The average communication traffic in the experiment was about 13.8 GB. As the number of clients expands to a larger scale, the computational load and scheduling complexity of the coordination layer will further increase. In addition, although the non-independent and identically distributed data partitioning used in the experiment can simulate real heterogeneous scenarios, the convergence efficiency of the model under extreme distributions still needs further verification. The above limitations will be optimized in future work by introducing asynchronous aggregation and lightweight communication protocols.

Security Analysis of IoT Traffic Classification Systems Under Adversarial Machine Learning Attacks

Chenxuan Li — 2026-08-09

A constraint-aware adversarially robust Internet of Things (IoT) traffic classification system with protocol validity, device behavior consistency, and manifold-aware training and evaluation is presented in this study. In realistic IoT communication semantics, resilience as a constrained min–max optimization problem allows adversarial perturbations. Comprehensive testing on sample IoT traffic datasets shows that baseline models achieve 95.1% accuracy under benign conditions but plummet following hostile attacks. The proposed defense reduces untargeted attack success rates to <18% while achieving 81.3% accuracy at ε=0.05 and 70.6% at ε=0.10. The proposed constraint-aware adversarial framework significantly enhances IoT traffic classification by achieving 97.4% accuracy and maintaining 90.6% robustness at ε=0.10, outperforming state-of-the-art methods. It reduces attack success rates to 11.2% (untargeted) and 7.9% (targeted) through protocol-compliant perturbations and manifold-aware learning. Additionally, the model achieves an efficient trade-off with 21.4 ms latency and 650 flows/sec throughput, making it suitable for real-time edge deployment. These results demonstrate improved robustness, realism, and deployability compared to existing approaches.

Enhancing Network Communication Security Using Hybrid Cryptographic Techniques

Qing He — 2026-08-09

The security requirements of networked systems have become increasingly critical due to the growing need for intelligent systems that can detect intrusions and protect data during transmission. This study presents a network security system that combines deep learning-based traffic trust assessment with two different cryptographic protection methods. The system employs a DenseNet–BiGRU design to capture network traffic patterns across different spatial and temporal dimensions, enabling the system to distinguish between normal and malicious traffic before the data is encrypted. The system uses Elliptic Curve Cryptography (ECC) to secure session establishment for trusted traffic, which enables key exchange and implements Advanced Encryption Standard (AES) for data encryption that requires low computational resources. The proposed framework reaches an accuracy of 94.5%, together with a precision of 88.8%, recall of 82.3%, F1-score of 85.4% and Matthews Correlation Coefficient (MCC) of 0.82, which demonstrates its ability to detect under conditions of class imbalance. The model demonstrates exceptional ability to differentiate between classes, which results in an ROC-AUC of 0.96 and PR-AUC of 0.93. The analysis of cryptographic performance shows that encryption and decryption process times remain minimal while system performance maintains consistent throughput, which increases with larger payloads. The framework demonstrates its ability to detect attacks in real time while maintaining secure communication, which makes it suitable for modern network protection and IoT security frameworks.

A Deep Semantic Confusion Vulnerability Detection Method Based on Tensor Recurrent Matrices and Gated Graph Convolutional Neural Networks

Yao Hu — 2026-08-09

Under the normalized network security situation of artificial intelligence-assisted attacks, deep semantic obfuscation has become the core means of vulnerability hiding. Attackers evade detection by legitimizing semantic associations and obfuscating code logic, posing serious threats to infrastructure and software supply chain security. Therefore, to enhance the robustness and accuracy of vulnerability detection under deep semantic obfuscation scenarios, the research proposes a vulnerability detection method based on tensor circulant matrix. The method preserves code semantic integrity and local correlations based on tensor circulant matrix. On this basis, it combines Gated Graph Convolutional Neural Networks (GGCNN) to improve the model’s feature capture capability for hidden vulnerabilities. On obfuscated vulnerability datasets, the average detection accuracy for obfuscated vulnerabilities reaches 96.24%, precision reaches 83.62%, recall reaches 87.53%, and F1 score reaches 85.54%. Compared with the Long Short-Term Memory Network (LSTM) baseline model, the False Negative Rate (FNR) has decreased by 16.73%, demonstrating significantly improved robustness under semantic obfuscation scenarios. The vulnerability detection model constructed in this study can effectively resist semantic obfuscation interference, provides a reliable technical approach for deep semantic obfuscation vulnerability detection, and has important practical significance for strengthening code security protection.

HMG-AID: Heterogeneous Internet of Things Intelligent Intrusion Detection Model Based on Multimodal Graph Attention

Wei Wu — 2026-08-09

With the increasingly complex heterogeneity of device types, communication protocols and data formats in the Internet of Things (IoT) environment, traditional intrusion detection (ID) models are difficult to effectively deal with dynamic threats. This paper proposes an intelligent IoT intrusion identification model HMG-AID for heterogeneous environments. The model realizes end-to-end ID by fusing multi-modal feature extraction, graph attention mechanism and dynamic trust evaluation. It uses one-dimensional Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) to extract spatiotemporal characteristics of traffic data in parallel, constructs heterogeneous device graphs and aggregates neighbor information through GraphSAGE, then introduces multi-head graph attention-weighted key nodes, and finally integrates zero-trust dynamic evaluation to optimize classification decisions. The F1 scores of the HMG-AID model on the three datasets of NSL-KDD, CICIDS2017, and ToN-IoT are 94.7%, 95.3%, and 94.9%, respectively, which are significantly better than the baseline model. Moreover, the F1 score retention rate under-5 dB noise interference is 85.6%, and the F1 score retention rate under counterattack attack is 88.9%. In addition to this, the model has a 93.6% F1 score retention rate when the device size increases to 10,000 nodes. The model effectively improves detection accuracy, environmental adaptability and decision interpretability through multi-level innovative design and provides a reliable theoretical basis and practical framework for heterogeneous IoT security protection.

Secure Transmission Algorithms for Multiplexing Information of Inter-Domain of Wireless Sensor Networks

Xiaopeng Yan — 2026-08-09

To enhance the anti-attack capability of inter-domain wireless sensor networks and ensure secure information transmission, a secure transmission algorithm for multiplexing inter- domain wireless sensor network information is proposed. This algorithm utilizes the inter- domain structure of wireless sensor networks to analyze channel transmission characteristics, integrating the network information transmission principles of multiplexing technology to construct a robust communication framework. The transmission model of inter-domain multiplexed information in wireless sensor networks is constructed. This model primarily relies on the HRC-MAC protocol, leveraging the characteristics of code division multiplexing to perform uplink spread spectrum and downlink modulation of information. It also incorporates the threshold method for information encryption and transmission control, which can improve the security of information transmission while reducing the fading loss of channel transmission. The experiment constructs a wireless sensor network environment and uses the algorithm for secure information transmission. The experimental results show that the channel transmission security capacity of the algorithm exceeds 1.5 bps/Hz, and the transmission security rate meets the standard. The transmission security factors of the channel within the domain exceed 0.917. The anti-attack performance of information encryption exceeds 90.2%. It can effectively disrupt the information sequence and significantly prevent tampering during information transmission.

Multi-Heterogeneous Power Data Security Protection in Smart Grid Based on Data Aggregation and Paillier Homomorphic Encryption Algorithm

Binyuan Yan — 2026-08-09

Multi-heterogeneous power data in smart grid refers to power data that includes multiple types, modalities, and sampling frequencies, such as user electricity consumption, equipment operation, and grid scheduling. The diverse and heterogeneous power data in the smart grid is related to the stable operation of the grid and user privacy. Without effective protection, it is easy to cause risks such as information leakage and scheduling failure. Therefore, targeted security protection solutions need to be constructed. However, there are problems with the loss of information granularity, high risk of privacy leakage, and limited data analysis in the current smart grid power data aggregation and sharing. To enhance the security protection effect of power data, a multi-dimensional data security protection scheme based on data aggregation and Paillier homomorphic encryption is proposed. Firstly, a three-tier system model for multivariate heterogeneous power data in smart grids is constructed (smart meters, data collection stations, and blockchain nodes). Subsequently, the Paillier homomorphic encryption algorithm is integrated to encrypt and aggregate users’ multi-dimensional electricity consumption data. At the same time, data aggregation and consortium chain technology have been introduced. Experimental results demonstrate that this scheme offers significant advantages over traditional Rivest-Shamir-Adleman (RSA) encryption schemes, traditional Advanced Encryption Standard (AES) encryption schemes, Elgamel encryption schemes, and traditional Transmission Control Protocol/Message Queuing Telemetry Transmission Protocol transmission schemes in terms of computational and communication overhead. When the number of users reaches 5000, the computational overhead at data collection stations is only 35.6% of that in traditional RSA methods, and the communication overhead is merely 28.7% of traditional transmission control protocol methods. Additionally, when transmitting power data from 5000 users simultaneously, the information accuracy rate exceeds 92%, and the packet loss rate remains below 0.5%. In conclusion, the proposed scheme provides an efficient and reliable technical pathway for the secure transmission of multivariate heterogeneous power data in smart grids.

Lightweight Deep Learning Optimization of AES for Mobile IoT Security

Yanxin Zhang — 2026-08-09

Mobile Internet of Things (IoT) terminals are typically constrained by limited computing power, memory, and battery life, making traditional implementations of the Advanced Encryption Standard (AES) suffer from excessive computational overhead, high latency, and insufficient resistance to side-channel attacks. To address these key limitations, this paper proposes a lightweight deep learning-based AES optimization scheme tailored for resource-constrained environments. We designed a lightweight Convolutional Neural Network (CNN) based on MobileNet depthwise separable convolutions to reconstruct the SubBytes transformation, and a lightweight Long Short-Term Memory (LSTM) network enhanced with structured pruning and 8-bit quantization to optimize the key expansion module. This approach significantly enhances the nonlinearity of encryption operations and the randomness of round keys while reducing model parameters and computational load. Extensive experiments on the ARM Cortex-M4 chip demonstrate that the optimized AES achieves a 38% improvement in encryption throughput and a 42% reduction in key expansion time. Security evaluations show that the round key Shannon entropy increases to 148.2 bits, with substantially enhanced resistance to differential attacks, linear attacks, and Differential Power Analysis (DPA, with Correlation Power Analysis [CPA] as its mainstream engineering implementation). Notably, the optimized algorithm only increases on-chip memory usage by 12% and reduces static power consumption by 18%, making it suitable for deployment on low-resource mobile IoT devices.