• 1: Formulates low-light enhancement as a counterfactual intervention on a Retinex SCM. • 2: Introduces a physics-guided structural causal model for illumination–reflectance reasoning. • 3: Achieves superior perceptual fidelity and robustness on multiple benchmark datasets. Many Architecture, Engineering, and Construction (AEC) operations must operate safely under visually compromised conditions, dimensional dimension, including night-time navigation/driving, tunnel or cave exploration, and hazardous or confined sites observation. In such conditions, low-light image enhancement is expected to be more than simply image brightening but also to maintain mission-critical structures and be visually interpretable for deployment in safety-critical conditions. Nonetheless, current deep LIE techniques conceptually overlook image processing, physical priors and causal mechanisms during improvement, contributing to constrained ruggedness and generalisability. This paper presents a Learning Probabilistic Low-light Image Enhancement (LPIE) network that explicitly integrates causal factors into the LPIE and allows for probabilistic counterfactual reasoning. LPIE casts illumination enhancement in the light of an intervention and unrolls it into a structured causal model of image formation. A normalizing-flow module performs invertible and physically coherent illumination processes, while an uncertainty-aware Transformer recovers reflectance in spatially varied and high-textured regions. Consequently, a complete probabilistic refinement further optimizes all elements to optimize the distribution of physical variables, resulting in a natural, fine-detailed photo and supporting across-used scenarios. Experiments on public benchmarks and AEC-relevant datasets show that LPIE achieves state-of-the-art performance and clearly surpasses recent LIE methods on the widely used benchmark dataset, paving the way for more interpretable, robust and trustworthy automated perception systems.

In hazy weather with high humidity, insulators in overhead transmission lines are prone to flashover. However, fog reduces the sensitivity of image detection, which is a commonly used method for condition monitoring of electrical equipment. Traditional image dehazing methods struggle to handle the complex backgrounds found in the scene and lack compatibility with downstream detection tasks. To overcome the image degradation, this study introduces Tongji Dehaze Network (TJDe-Net), a dehazing network engineered to enhance the visual clarity of images captured by unmanned aerial vehicles (UAVs) during power equipment inspections in hazy conditions. This network leverages a Swin-Transformer architecture with deep recursive feature integration and a cubic attention mechanism, therefore mitigating atmospheric degradation effects. Another major advancement presented in this work is the creation of the TJDehaze dataset, a comprehensive collection of paired images specifically designed for power transmission domain. This dataset comprises both real and synthetically generated hazy images, crafted to mimic a wide range of atmospheric densities and challenges. TJDe-Net’s performance is thoroughly evaluated across multiple datasets, including real-world dataset, TJDehaze synthetic dataset and the SFID-improved dataset. These evaluations proved TJDe-Net’s superior dehazing efficacy, which significantly bolsters subsequent image detection tasks. Experimental outcomes confirm TJDe-Net’s robustness and adaptability in real-world scenes, thereby enhancing the reliability and efficiency of UAV-based inspections and suggesting extensive potential for applications requiring enhanced visual perception in adverse weather conditions.

Cellular backhaul and core networks have traditionally been considered as Walled Garden, with their security ensured by physical isolation. Therefore, prior security studies primarily focused on radio access networks with limited treatment of backhaul and core network interfaces. In this paper, we performed a security evaluation of real-world GPRS Tunnelling Protocol (GTP) deployments. GTP is the fundamental protocol for user traffic management between base stations and core networks (inside the Walled Garden) from 3G to 5G, thus often assumed inaccessible and non-exploitable from the Internet. However, our study reveals for the first time the troubling state of GTP access control in real-world deployments. Aided by a semi-automated tool, our measurements discovered around 749,000 valid GTP hosts accessible via the public Internet, spanning across 1,176 service providers in 162 countries. Our results demonstrate potential exposure of mobile core network infrastructures to external threats. We then evaluated the attack surface of exposed GTP infrastructures, and found out that as many as 38 types of GTP messages can be misused to launch various attacks such as denial-of-service and session hijacking. Our experiments using open source 4G and 5G projects in isolated lab environments further confirm the feasibility of those GTP-based attacks, including remote hijacking of user traffic sent through cellular core networks. In addition to threats against cellular networks and their subscribers, exposed GTP devices could also be weaponized to launch large-scale reflective denial-of-services (RDoS) attacks. We hope our findings will increase awareness of GTP vulnerabilities among operators and the security community, highlighting the urgent need to further strengthen security in cellular core networks.

The development of 6G should enable users in remote and harsh areas to enjoy computation-intensive services including metaverse entertainment, intelligent transportation, and immersive communications. Low Earth Orbit (LEO) satellite constellations widely constructed in recent years have been recognized as an efficient solution to complement the terrestrial infrastructure with seamless coverage and decreasing expenses for both communication and computation services. However, the widely studied Federated Reinforcement Learning (FRL) based task offloading strategies neglect the potential trust concerns like malicious satellites and buffer pollution, while 6G service providers may rent the LEO satellites belonging to different companies to minimize the expense. To address these issues, blockchain has been considered in the Zero Trust (ZT) scenario, with the group consensus mechanism through the smart contract. Moreover, we propose a Constrained Correction Voting Mechanism (CCVM) to give punishing correction to the aggregation weight of malicious voting satellites. Furthermore, a Cold Start Reputation Aggregation (CSRA) scheme is adopted to first severely degrade and then gradually recover the weight of Federated Learning (FL) sub-models trained by malicious satellites. Thus, the Blockchain-enabled Cold Start Aggregation FRL (BCSA-FRL) scheme is proposed to make effective and secure offloading decisions in the ZT LEO satellite Networks. The numerical results illustrate the advantages of our proposal.