Electronic accounting voucher deposits and cybersecurity audits face a contradiction between resistance to evidence tampering and data privacy protection. Using blockchain technology as a basis, the authors propose a solution framework that addresses both concerns. Under a dual-layer hash mapping architecture, original accounting vouchers are stored off-chain via the InterPlanetary File System, and digital fingerprints are anchored on the immutable ledger. Smart contract engines enforce a multi-signature state machine to ensure accounting voucher traceability from generation to archiving. Paillier homomorphic encryption supports direct computation on ciphertext, enabling continuous consistency checks between accounts and accounting vouchers without decrypting raw data. Experimental results showed that on-chain storage overhead was reduced by 99.08%, audit latency reached 12.4 s under a workload of 5,000 transactions per second, and no raw data were exposed throughout the process. This framework provides forensic-ready support for secure electronic accounting voucher management.
Law enforcement tactical training involves high-pressure scenarios, but traditional methods are limited by difficult scene simulation, incomplete records, insufficient quantitative evaluation, and delayed feedback. Training quality relies heavily on instructors’ experience, leading to inconsistent standards and poor repeatability, especially in multi-agent collaboration and rapid decision-making. This study uses artificial intelligence (AI) to build a full-process closed-loop framework for scenario generation, behavior collection, evaluation, feedback optimization, and retraining. Results confirm AI significantly improves training efficiency, feedback speed, collaboration quality, and capability assessment accuracy. The framework forms a verifiable, replicable training paradigm, and full-process data can serve as standardized digital evidence for law enforcement review and supervision.
This paper proposes a formally verifiable access control framework that integrates structured eXtensible access control markup language policy specification with finite-state abstraction and symbolic model checking using NuSMV. To evaluate the proposed framework, a simulated 150-node Internet of Things-based critical infrastructure environment was developed in MATLAB. The framework models three operational states, namely normal, abnormal, and emergency, while incorporating a contextual risk threshold of 0.8 for role-based authorization involving operator, emergency responder, adversary, and policymaker roles. Access control policies were transformed into a finite-state transition system and formally verified for safety, liveness, and non-interference properties. Experimental validation across five consecutive verification runs confirmed that all specified properties were satisfied without counterexamples. Performance evaluation demonstrated an average authorization latency of 0.154 ms, mean CPU utilization of 58%, and average energy consumption of 1.5 J.
Traditional methods struggle to detect concealed environmental crimes. This study proposes an intelligent framework that uses machine learning and multi-source data fusion (environmental, operational, and logistics) to proactively identify risks such as illegal discharge and waste transfer. Empirical results show improved timeliness and accuracy of detection. Despite high initial costs, long-term economic benefits are substantial. Challenges remain in algorithmic transparency, data sharing, and legal compliance. The framework provides actionable intelligence for law enforcement, bridges data silos to enable coordinated responses, and contributes to digital forensics and sustainable governance. Future integration with blockchain technology could enhance the integrity of digital evidence for prosecution.
Large vision language models (VLMs) show strong open-world generalization but degrade at domain-specific tasks, while traditional small forensic models perform well on in-distribution datasets yet lack cross-distribution generalization and language-based interpretability. To address this gap, the authors propose a vision forgery trace (VFT)-VLM framework, which incorporates forensic features into a VLM without sacrificing its general reasoning ability. Specifically, a lightweight VFT extraction module learns to encode texture anomalies, edge incoherence, pixel artifacts, and frequency-domain deviations. The traces are incorporated into the InternVL2-8B backbone via low rank adaptation fine-tuning, achieving alignment between visual evidence and textual explanations. Across 14 diverse artificial intelligence-generated content benchmark datasets, VFT-VLM outperforms VLM-based large-scale models and achieves comparable or superior performance to relevant traditional small-scale models. Ablation studies confirm both VFT extraction and low rank adaptation fine-tuning are critical to the performance gains.
Despite growing development of digital forensic tools for detection of child sexual exploitative and abuse material (CSEAM), victims and offenders remain a challenge to investigators and forensic experts. To understand developments and shortcomings of digital forensic approaches, a systematic literature review was carried out in IEEE Xplore, EBSCOHost Academic Search complete, and Science Direct between 2010 to 2025. A total of 41 articles out of 4,148 were selected through various filtering criteria. The review revealed seven themes covering the dark web, detection tools, crime patterns, applications based on artificial intelligence (AI) and machine learning (ML), biometric analysis, social media network analysis, and analysis of online behaviour. Despite the growing popularity of AI and ML, their application towards addressing CSEAM is scanty. Text analysis is the least commonly used feature, though text accompanies all media. Ethical implications are discussed. This research will help relevant stakeholders to strengthen the fight against CSEAM.
To combat evolving Android malware, this paper proposed a lightweight deep learning detection system leveraging Drebin, AndroZoo Lite, and Canadian Institute for Cybersecurity MalDroid 2020 datasets. The approach fused static (permissions, call graphs) and dynamic (behavior logs) features via a hybrid convolutional neural network-Transformer model with bidirectional information flow for end-to-end training. To meet mobile device constraints, joint optimization through network pruning, quantization, and attention distillation was applied. Evaluated via five-fold cross-validation, the method outperformed baselines (support vector machine, long short-term memory, BERTroid (BERT-based Android Malware Detection Model), convolutional neural network-Vision Transformer) in precision, recall, F1, area under the curve, and inference latency, achieving high accuracy with low delay. It remains robust against polymorphic and obfuscated variants. Error analysis reveals the critical impact of feature fusion weights on decision-making, offering insights for real-time mobile threat defense.
With the rising prevalence of telecom fraud, identifying key factors influencing deterrence has become essential for enhancing anti-fraud strategies. This study adopts a victim-centered approach and examines 35 representative cases using a triadic "actor-environment-interaction" framework. Applying Necessary Condition Analysis and fuzzy-set Qualitative Comparative Analysis, it uncovers multiple configurations associated with successful deterrence. Results indicate that deterrence effectiveness does not rely on a single condition but emerges from complex causal combinations. Intervener professionalism and timely response are central in several pathways, while factors like case complexity and victim susceptibility also shape outcomes under certain conditions. Grounded in Complex Adaptive Systems theory, the study highlights the dynamic, non-linear nature of fraud deterrence and provides evidence-based recommendations for precision-oriented intervention strategies.
Victimization extends criminal cases to more complicated and volatile harm due to various forms of vulnerability and behavioral risk factors of victims. Multiple tools have been developed to identify victims, but few have been developed to assess, predict, and prevent potential victimization using machine learning. The objective of this research is to develop novel methods that aid in identifying potential victims to prevent crime. This paper proposed a prediction of victimization using a mixed ML/DL approach, based on a self-administered dataset of 880 individuals. The data recorded personal and behavioral characteristics. Missing data were handled, and the ML algorithms were assessed after normalization. The authors utilized several machine learning and deep learning classifiers, which were selected due to their applicability to structured survey data and their ability to model non-linear relationships flexibly. For performance evaluation, they utilized nine models. The results indicated that the K-Nearest Neighbors gained a high accuracy of 97.73% and performed well compared to other models.
This study focuses on the technical concealment and fixed evidence of copyright infringement on online education platforms and puts forward an analysis framework including “technical feature deconstruction-infringement identification-criminal regulation path” and a binary judgment standard of “technical intrusion intensity-subjective cognition degree.” By generating a countermeasure network (GAN) to detect the anomaly of user behavior logs, the technology-neutral boundary that conforms to the modesty of criminal law is established, and the crime identification rules based on the reverse cracking of encryption technology, the accomplice evaluation model with distributed storage characteristics, and the criminal-criminal cooperation mechanism with smart contract code review as the core are formulated. Experiments show that the matching degree of elements of the framework reaches 98.2% in a single scene, but the technology-neutral evaluation module needs to be optimized in mixed scenes, which provides a systematic solution with both theory and practice for online education copyright protection.
Robust weapon detection from drones has drawn much attention from security and law enforcement communities, such as first-response drones. Multiple deep-learning techniques have been developed to detect weapons in the context of public safety. However, these techniques rely on the datasets gathered from YouTube videos, CCTV cameras, and other internet resources, which are difficult to generalize for drone-based detection. In this work, the authors investigate the limitations of existing datasets and models for drone-based applications and re-train the models (YOLOv5, v7, and v10) using drone-collected datasets (color and infrared images) in three different environments: laboratory, car park, and desert environments. Real metallic guns, pistols, and knives are used in the experiments. Experimental results show that YOLOv5 outperforms the existing models, achieving 91.4% precision for the indoor infrared spectrum.
With the rising prevalence of telecom fraud, identifying key factors influencing deterrence has become essential for enhancing anti-fraud strategies. This study adopts a victim-centered approach and examines 35 representative cases using a triadic “actor–environment–interaction” framework. Applying Necessary Condition Analysis and fuzzy-set Qualitative Comparative Analysis, it uncovers multiple configurations associated with successful deterrence. Results indicate that deterrence effectiveness does not rely on a single condition but emerges from complex causal combinations. Intervener professionalism and timely response are central in several pathways, while factors like case complexity and victim susceptibility also shape outcomes under certain conditions. Grounded in Complex Adaptive Systems theory, the study highlights the dynamic, non-linear nature of fraud deterrence and provides evidence-based recommendations for precision-oriented intervention strategies.
The increasing sophistication of phishing attacks poses a significant challenge to cybersecurity. Modern phishing tactics utilize advanced anti-detection mechanisms, resulting in substantial economic losses. Attackers often meticulously replicate well-known brand websites to exploit user trust and maximize financial gain. Current detection methods struggle to differentiate between these imitations and legitimate websites due to difficulties in extracting essential website features and underutilization of multimodal data. As a result, existing approaches fail to meet the practical demands of robust security protection. To address these limitations, this paper introduces CMNet, a novel approach that leverages contrastive learning to identify subtle differences between phishing and legitimate websites. CMNet effectively integrates multimodal website data to generate a comprehensive feature representation. Evaluated on the largest publicly available phishing dataset, CMNet achieves an accuracy of 96.3%, demonstrating superior performance.
This article examines the pivotal role of big data in transforming psychological prevention models for juvenile delinquency. With the rapid advancement of information technology, big data offers a new lens through which to understand the criminal psychology of minors. Through the systematic collection and analysis of relevant data, it becomes possible to more accurately identify the psychological traits and potential risk factors associated with juvenile delinquency. The application of big data has facilitated a shift from traditional, experience-based approaches to more scientific, data-driven precision prevention. This transformation not only enhances the effectiveness and specificity of intervention efforts but also provides stronger support for the healthy development of adolescents. In summary, big data presents both new opportunities and challenges for psychological prevention and holds significant promise for improving overall prevention outcomes.
This article focuses on the legal regulation of price discrimination against existing customers using big data. This practice infringes on consumers' rights to know, fair trade, and personal information protection. However, the current legal system has flaws. For example, price discrimination lacks clear definition and punishment, and there's no accurate definition of “legitimate reasons”. Also, current legislation has obvious gaps in preventing algorithm - related risks. To improve the legal regulation, it is recommended to incorporate differential pricing for consumers into the scope of adjustment of the Price Law, refine the provisions on fair trade conditions in the Law on the Protection of Consumers' Rights and Interests, and improve the rights relief mechanism. Moreover, consumers should be more aware of rights protection, be more price - sensitive, keep evidence, and use judicial procedures or consumer associations to safeguard rights. We aim to create a healthy and orderly Internet economy, ensuring consumers' equal treatment and legal rights in digital transactions.
Legal Named Entity Recognition (NER) is crucial in intelligent judiciary systems, focusing on identifying case-specific entities in legal texts. It helps convert unstructured legal documents into structured data, improving e-discovery efficiency. However, challenges arise from insufficient understanding of legal terminology, leading to errors in identifying long and nested entity boundaries. To address this, a Legal NER method based on a parallel instance query network is proposed. This method uses learnable instance queries to extract entities in parallel, with a BERT+BiLSTM+attention structure to encode context and query information. Entity prediction is performed using a pointer network to identify span boundaries and entity types. A linear label assignment mechanism aligns legal entities with queries for more accurate labeling. Experimental results show that the model outperforms existing methods, and further validation through ablation experiments and case studies supports its effectiveness, offering valuable insights for advancing legal NER research.
Detecting dangerous objects, such as firearms or knives, is crucial for public safety or accurate situational assessment in crime scenes in law enforcement applications. Drones as first responders have been actively utilized for this purpose, showing significant benefits in law enforcement with fast and early detection of such objects. However, automated detection is still challenging, particularly with low-quality drone cameras that operate in low illumination conditions. We evaluate the performance of four popular AI deep learning models to automate the detection of dangerous objects recorded from low-quality drone cameras. The results show that the YOLOv5s model achieves the best detection performance, yielding mAP50 results of 0.964 for color and 0.949 for infrared videos, which are excellent performances considering the low-quality and low-resolution dataset. The trained network model is further implemented as an online web application where law enforcement officers can upload videos taken from drones or CCTV.
Mobile phones and computers are widely used devices these days, with almost everyone carrying a smartphone and multiple personal computing devices at their homes. Unfortunately, the perpetrator exploits these devices for their unlawful activities. They employ various tactics such as sending phishing emails, and malicious links to harvest confidential information and exploit users. The perpetrators often leave traces on search engines, where they search for illegal materials and weapons, or send threatening emails to victims. This paper primarily focuses on locating and retrieving browsers' artifacts while considering the challenges posed by private browsing modes, which perpetrator may use to cover their tracks. The study also compares well-known search engines like Edge, Safari, and Firefox, analyzing the strengths and weaknesses of their directories. Moreover, it explores evidence extraction from smartphones, comparing the success rates between rooted or jailbroken phones and evidence obtained from browsers versus applications.
In vechcular networks, a promising approach to enhance vehicle task processing capabilities involves using a combination of roadside base stations or vehicles, there are two challenges when integrating the two offloading modeth: 1) the high mobility of vehicles can easily lead to connectivity interruptions between nodes, which in turn affects the processing of the tasks that are being offloaded; and 2) vehicles on the road are not completely trustworthy, and vehicle tasks that contain private information may suffer from result errors or privacy leakage and other problems. This paper investigates the computing offloading problem for minimizing task completion delay in vehicular networks. Specifically, we design a trust model for mobile in-vehicle networks and construct a migration decision problem to minimize the overall delay of task execution for all vehicle users. The simulation results show that the scheme proposed in this paper can effectively reduce the execution delay of the task compared to the baseline scheme.
The traditional laboratory anomaly detection methods mainly focus on the hidden dangers caused by chemical leaks and other items, ignoring the impact of abnormal behaviors such as incorrect operations and improper behavior on safety in the laboratory. This paper proposes a laboratory abnormal behavior detection method based on multimodal information fusion. The method generates a dense optical flow field of RGB image sequences based on optical flow theory and global smoothing constraints, and mines motion mode information. Meanwhile, the contour modal information of behavior is captured through convolution and adjacency matrix operations. Using decision level and proximity functions to integrate student behavior motion mode information and contour mode information, and using the maximum value as the behavior detection result. The experimental results show that the method can effectively detect abnormal behavior in the laboratory environment, with small detection errors and a specificity close to 1.00, effectively ensuring the safety of the laboratory environment.
