Table of Content

24 July 2026, Volume 12 Issue 7

Previous Issue   

Survey of Software Supply Chain Security Detection and Assessment Technologies

2026, 12(7):  586-597. 

Asbtract ( )   PDF (1749KB) ( )  

References | Related Articles | Metrics

In the context of the digital era, software supply chain has become a critical component supporting the stable and healthy development of the digital economy. It is an indispensable part of the nation’s key information infrastructure and economic and social systems. The security of software supply chain directly determines the security of the key businesses carried by the software supply chain. Therefore, based on the development needs of the digital age, this article summarizes the current technologies, methods, and development trends related to software supply chain security detection and evaluation, providing reference and guidance for industry insiders, researchers, and decisionmakers. It includes a review and detailed explanation of the background and methods of software supply chain security detection and evaluation technology, detailing the principles of mainstream technologies such as component analysis, vulnerability scanning, code review, runtime monitoring, threat modeling, and fuzz testing, and comparing and analyzing the advantages and disadvantages of various technologies; Analyze the current technical challenges and countermeasures faced by technology; And propose ten major trends for the development of this field in the next decade, in order to improve the security level of the software supply chain and promote the development of the software industry.

Interactive Dynamic Privacy Risk Assessment and Adaptive Protection Methods in Data Publishing

2026, 12(7):  598-605. 

Asbtract ( )   PDF (1785KB) ( )  

References | Related Articles | Metrics

Data publishing is an important way to promote data sharing. However, studies have shown that it is accompanied by the risk of privacy leakage of sensitive attribute information due to large and frequent access to data by malicious parties. Although existing evaluation methods consider the scenarios of above issue, most of them are noninteractive and static settings. Therefore, in practical applications, dynamic risk cannot be accurately assessed by data publishers, and even the noise protection causes an imbalance between privacy and utility. In this paper, we propose an interactive dynamic privacy risk assessment and adaptive protection method. It firstly utilizes machine learning prediction models to determine the privacy level of difficulttodefine attributes. And secondly we sets a mechanism to dynamically adjust the privacy level according to the actual requests of the data demander and timely restricts the leakage of sensitive attribute caused by highfrequency access. Finally, an adaptive noiseadding mechanism is proposed after obtaining the high risk of evaluation feedback, to ensure that the balance of privacy and utility achieveing good results when the data is released. Experimental results show that the privacy risk index increases gradually with the number of accesses and the accesses containing associated attributes lead to an increase in the privacy risk index by more than 15%. Adaptive noise addition, on the other hand, provides a similar level of privacy protection to overall noise addition at 0.5, yet improves data utility by more than 30% over overall noise addition.

The Mechanism of Disinformation Generation and Governance Pathways in Generative AI Models

2026, 12(7):  606-612. 

Asbtract ( )   PDF (1255KB) ( )  

References | Related Articles | Metrics

While driving transformations in online information order, generative AI models also generate disinformation risks characterized by an “objective+subjective” overlap. An analytical framework tailored to their technical characteristics is urgently needed. This study systematically deconstructs the technical logic of false information production in generative AI models based on their hybrid expert architecture, treelike reasoning patterns, localized semantic understanding attributes, and opensource ecosystem mechanisms. It analyzes the transmission mechanisms of false information risks across four stages: data input, algorithmic operation, content presentation, and cognitive dissemination. To address these risks: At the algorithmic level, implement a processbased oversight scheme encompassing “access reviewoperation disclosurepostevent verification”; at the presentation level, strengthen scenariobased, interactive “warning notice” labeling mechanisms; at the cognitive level, cultivate users' digital literacy and selfrestraintt capabilities to achieve effective information security governance in the AI era.

A Deep Learningbased Method for Background Traffic Generation in Railway Cyber Range

2026, 12(7):  613-624. 

Asbtract ( )   PDF (2995KB) ( )  

References | Related Articles | Metrics

Cybersecurity threats have permeated all aspects of the railway industry, necessitating that railway cybersecurity ranges keep pace with the latest security demands. Background traffic is a core element in constructing a realistic simulation environment for cyber ranges and serves as a critical technology supporting range operations. However, effectively capturing the complex spatiotemporal characteristics of network traffic and generating highfidelity background traffic remains a significant challenge. This paper proposes a novel method for generating background traffic in railway cyber ranges (referred to as B2Diff). By integrating deep learning techniques such as BERT, bidirectional long shortterm memory networks, and diffusion models, the method captures features such as contextual semantics, spatiotemporal dependencies, and complex traffic distributions in traffic samples, thereby generating highly realistic and diverse background traffic. Experimental results demonstrate that B2Diff significantly outperforms existing methods in terms of the quality and diversity of the generated background traffic, validating its effectiveness in enhancing the simulation realism of cyber ranges.

Multidomain Fake News Detection Model Based on Prompt Learning and Fuzzy Labels

2026, 12(7):  625-633. 

Asbtract ( )   PDF (1229KB) ( )  

References | Related Articles | Metrics

The widespread popularity of the Internet and intelligent devices has provided convenience for the public to access news. However, this also creates a breeding ground for the generation and propagation of fake news. Fake news spans multiple domains, whereas existing detection models often overlook the specificity of corpora across different domains, limiting their accuracy. To adress this issue, this paper proposes a multidomain fake news detection model based on prompt learning and fuzzy labels. This model employs RoBERTa to extract textual features and reformulates the detection task as a cloze problem by constructing prompt templates containing domain characteristics; meanwhile, it utilizes domain fuzzy membership probabilities generated by a neural network to guide the prompt learning process, effectively enhancing accuracy and generalization ability. Experimental results on the public datasets Weibo17 and Weibo21 demonstrate that this model outperforms traditional finetuning methods and existing stateoftheart methods under both domainunlabeled and multidomain conditions, with an average F1 score improvement of 1.16 percentage points, validating its feasibility and effectiveness in multidomain fake news detection tasks.

Deepfake Face Detection Method Based on Multiloss Fusion

2026, 12(7):  634-643. 

Asbtract ( )   PDF (2941KB) ( )  

References | Related Articles | Metrics

Most existing deepfake face detection methods rely on specific forgery patterns, such as noise artifacts or local textures, making them highly dependent on known forged features and lacking generalization to unknown forgeries. To address this issue, this paper proposes a multiloss fusion detection framework based on the reconstructionclassification learning(RECCE) network. A prototype similarity mechanism is introduced to measure the distance between sample features and class prototypes, enhancing the model’s ability to detect unknown forgeries. A joint loss function combining prototype loss, binary crossentropy loss, reconstruction loss, and metric learning loss is designed to strengthen feature learning from multiple perspectives. Moreover, multilevel encoder features are fused, and a reconstructionguided attention mechanism focuses the model on forged regions rather than the entire face, improving robustness and accuracy. Experiments conducted on several benchmark datasets and compared with six stateoftheart methods demonstrate significant improvements: training time is reduced to 28% of the original model, and the AUC increases by 1.48% in crossdomain evaluation when trained on FaceForensics++ (c40) and tested on CelebDF. The results verify the superior performance and generalization ability of the proposed method.

A Multistrategy Data Augmentation Approach for Human Activity Recognition

2026, 12(7):  644-651. 

Asbtract ( )   PDF (1990KB) ( )  

References | Related Articles | Metrics

Wearable human activity recognition (HAR) plays an important role in intelligent health monitoring and behavioral analysis. However, realworld applications often encounter challenges such as limited user diversity and insufficient labeled data, which constrain the model’s generalization capability. To enhance crossuser adaptability while preserving data privacy, this study proposes a privacyfriendly human activity recognition model based on multistrategy data augmentation. The proposed framework integrates source pretrained networks with targetdomain data to achieve domain adaptation without requiring access to the source data. In the source domain, multiaxis sensor signals are transformed into activity images, and spatialtemporal representations are extracted using a GCNTransformer backbone for pretraining. In the target domain, multiple data augmentation strategies are employed to generate diverse views of activity samples. Furthermore, threelevel consistency constraints, applied to intermediate features, logits outputs, and pseudolabel confidence, enhance the model’s domain adaptability and robustness. Experimental results demonstrate that the proposed model achieves superior accuracy and generalization across users, providing a promising solution for personalized and privacypreserving applications in wearable computing.

Image Steganography Model Based on Improved Generative Adversarial Network and Selfdistillation

2026, 12(7):  652-661. 

Asbtract ( )   PDF (2834KB) ( )  

References | Related Articles | Metrics

Existing image steganography methods have made significant progress in concealment and antiattack capabilities, but still suffer from low image quality, limited information embedding capacity, and insufficient model stability and generalization ability. To address these issues, this paper proposes an image steganography model based on an improved GAN and selfdistillation (RCSDGAN). First, a residualchannel attention mechanism is designed and integrated into the DenseNet architecture to optimize its structure. Second, an encoding network and decoding network based on the enhanced DenseNet are constructed, combined with a discriminative network to form a complete image steganography framework. Finally, a selfdistillation training strategy is introduced. The selfdistillation loss is defined by calculating the difference between the teacher model’s output and the student model’s output from the same network, and this loss is incorporated into the overall loss function to enhance model stability and generalization capability. Experimental results demonstrate that, at an embedding rate of D=1bpp, the steganographic images achieve a PSNR of 50.4131dB, an SSIM of 0.9992, and an Accuracy of 99.96%.

A Unified Distributionpreserving Residual Watermark Embedding Method for Diffusion Models

2026, 12(7):  662-671. 

Asbtract ( )   PDF (2732KB) ( )  

References | Related Articles | Metrics

Aiming at the challenge that diffusion model watermarking can hardly achieve both generation quality preservation and antidistortion robustness, this paper proposes a unified distributionpreserving residual embedding method. Without modifying the diffusion backbone parameters, the proposed method adopts a twostage strategy. First, latent variables are mapped to a uniform distribution using the cumulative distribution function of the Gaussian distribution, and watermark bits are embedded via interval partitioning and inverse mapping, ensuring that the latent distribution remains unchanged. Second, a lightweight residual module is inserted into the midblock of the UNet, where watermark bits are projected and injected into feature maps. Only this module is trained, achieving watermark embedding with minimal extremely small perturbations. The training phase jointly optimizes diffusion reconstruction loss and watermark prediction loss. Experiments on Stable Diffusion 2.1 demonstrate that the method maintains high visual fidelity of watermarked images, and the perturbation amplitude is far below the perceptual threshold. Under various distortions scenarios including JPEG compression, cropping, Gaussian noise, brightness and contrast adjustments, color quantization, and Diffwa(diffusion models for watermark attack) reconstruction, bit accuracy consistently exceeds 90%. Particularly, nearperfect accuracy is achieved under distortionfree and Diffwa reconstruction conditions, effectively balancing generation quality and robustness.

Advanced Persistent Threat Detection Based on Generative Subgraph Contrastive Autoencoder

2026, 12(7):  672-682. 

Asbtract ( )   PDF (2019KB) ( )  

References | Related Articles | Metrics

With the increasing sophistication and stealth of cyber attacks, particularly the continuous evolution of advanced persistent threats (APT) targeting critical information infrastructure, which are characterized by high concealment and longterm persistence, accurately distinguishing intrusions from normal behavior has become a critical challenge. Provenancebased intrusion detection systems can capture finegrained causal relationships among system entities, demonstrating strong advantages in distinguishing benign from malicious behaviors and uncovering stealthy attacks. However, existing learningbased approaches still suffer from the absence of proper node weighting, insufficient utilization of edge features, and inadequate learning of local subgraph structures, while also facing high computational costs when applied to largescale datasets. To address these limitations, we propose GSCAE, a novel APT detection framework based on a Generative Subgraph Contrastive Autoencoder. First, we construct node representations by integrating edge interaction features with local clustering coefficients and compute node importance using the entropy weight method. Then, we design a generative subgraph contrastive learning algorithm that jointly incorporates edgelevel and topological losses to more effectively learn local structures and interaction patterns. Finally, the learned graph embeddings are fed into a lightweight node classifier to perform anomaly detection, achieving a balance between detection accuracy and computational efficiency. Experiments conducted on the DARPA public dataset demonstrate that GSCAE outperforms most existing learningbased approaches in both accuracy and efficiency, validating its effectiveness and practicality in complex host environments.