In the era of the digital and intelligent society, artificial intelligence technology has developed rapidly. Against this backdrop, deepfake technology has also undergone iterative upgrades, achieving a phased leap from early encoding-decoding networks to text-to-video large models. Deepfake has officially entered the 3.0 era, characterized by multi-modality, low thresholds, and ultra-realism, and it has already posed a severe threat to national security, public social order, the legitimate rights and interests of citizens, and even the security of judicial procedures. Therefore, the risk regulation targeting the abuse of deepfake technology is already urgent.Based on this, this paper will, on the basis of conducting an exploration and analysis of the crime types and harmful effects of artificial intelligence deepfake, combine relevant domestic and international literature and practical cases to explore the construction of a multidimensional risk governance framework integrating technology, ethics, and law into one, with a view to achieving effective regulation of deepfake risks in the new era.

As an important carrier of internet of things technology, intelligent wearable devices are profoundly changing people’s health management methods. Design an intelligent wearable embedded system, with the STM32F103C8T6 microcontroller as the core, integrating detection modules such as the MAX30102 heart rate sensor, and construct a physiological parameter monitoring and motion recognition system. The system achieves multi-sensor data fusion through Kalman filtering algorithm, completes motion pattern recognition using support vector machine classifier, and optimizes power consumption control through dynamic voltage frequency scaling technology. Experimental results show that heart rate detection accuracy reaches 96.8%, motion state recognition accuracy is 90.6%, system average power consumption is 18.5 mA with continuous operation exceeding 74 h, laying a technical foundation for the industrialization of intelligent wearable devices.

Amid the new wave of technological transformation driven by Artificial Intelligence (AI), data centers have emerged as critical strategic infrastructure underpinning algorithm training, large-scale data processing, and the provision of global digital services. This article offers a systematic analysis of the structural challenges confronting U.S. data centers under mounting pressures, including rapidly growing demand for computing power, constraints on energy and water resources, vulnerabilities in the supply of critical minerals, labor shortages, and regulatory and permitting bottlenecks. By examining these interrelated pressures, the study provides forward-looking insights into the systemic risks and governance requirements faced by data centers as essential infrastructure in the era of AI, thereby contributing policy-relevant perspectives for future infrastructure planning and regulation.

Aiming at the problem of high energy consumption caused by using fixed schedule control equipment in high-rise office buildings, the research on energy-saving control methods under the environment of Internet of Things is carried out. A four-layer Internet of Things architecture is constructed, and various types of sensors are deployed to collect energy consumption data in real time. A Bayesian network combined with attention mechanism is used to build an energy consumption prediction model. Based on the prediction results and multidimensional data, the hierarchical intelligent control strategy is designed, and the closed-loop feedback of abnormal energy consumption monitoring is established to realize the intelligent energy-saving control of high-rise office buildings. Experiments show that this method has high prediction accuracy and low energy consumption, and significantly improves the level of building energy saving.

Open-source Artificial Intelligence (AI) has emerged as a pivotal force in the development of AI. Breakthroughs such as DeepSeek have prompted significant industry reflection and offer valuable reference points. This article the progression of open source, delves into the technical merits and possible hazards of open-source AI, and examines the interplay between open and closed paradigms in AI development. In the context where technology and policy are intertwined, this research offers valuable perspectives for comprehending and steering the direction of AI progress while striking a balance between innovation and security.

To address signaling congestion and protocol heterogeneity in building weak current systems, this study proposes a distributed architecture based on “Device-Edge-Cloud” collaboration. The architecture utilizes an edge Deterministic Finite Automaton (DFA) to achieve normalized parsing of heterogeneous physical protocols. It employs a computation offloading mechanism based on the M/M/1 queuing model and a Lagrange optimization algorithm with a latency penalty term to control system long-tail latency. Additionally, a Kalman filter algorithm is applied to smooth network jitter and resolve audio-visual synchronization drift. Experimental results demonstrate that under 500 QPS high concurrency, the proposed architecture keeps P99 latency within 210 ms. Under weak network conditions, lip-sync error is limited to the 
±40 ms range. Furthermore, during WAN interruptions, the system ensures the deterministic execution of critical commands via an edge software-hardware dual redundancy mechanism. This work provides a reference engineering paradigm for constructing highly stable smart building systems.

Aiming at the problems of low recognition accuracy and poor scene adaptability of wildlife in forest and grass informatization monitoring, a two-stage “binary classification detection - species characteristic matching” architecture is proposed. The first stage adopts improved YOLOv8 for binary classification of wildlife and non-wildlife, enhancing the detection capability of small and occluded targets in complex scenes by introducing Convolutional Block Attention Module (CBAM) mechanism and optimizing Complete Intersection over Union (CIoU) loss function. The second stage constructs a species feature database based on ArcFace, achieving accurate recognition of crested ibis, siberian tiger and other species through cosine similarity calculation. Experiments on a forest and grass area dataset with more than 80 species and 
120 000 images show that the overall accuracy rate reaches 93.1%, which can adapt to complex backgrounds, light changes and occluded scenes, providing technical support for wildlife monitoring and protection.

Zero-shot robotic grasping focuses on realizing high-dimensional perception-to-control mapping under unsupervised semantic input. This paper analyzes a specific production line case of an intelligent sorting enterprise for automotive parts, and investigates the hierarchical Vision-Language-Action (VLA) architecture DexGraspVLA+, including its high-level semantic planning and low-level control mechanisms. A cross-modal alignment method and contact force embedding optimization strategy are proposed to construct a closed-loop system for grasp execution path generation and contact force regulation driven by semantic mapping.

With the increasing intensity and depth of mining, the difficulty of mining and the risk of disasters are increasing. Especially in mines with complex geological environments, once the risk monitoring is not improper, it will often directly induce various mine disasters. Limited by the problems of limited range, insufficient sensitivity and high degree of data discretization in traditional monitoring technology, it is often difficult to accurately and real-time monitor the disaster risk and its causative effects in the whole process of mining. In order to solve the above problems, this manuscript proposes a dual-drive mine disaster risk monitoring method that integrates optical fiber sensing and data retrieval. Firstly, the vertical and horizontal combined optical fiber neural perception network is constructed to perceive the disaster risk of mining engineering in real time. Secondly, based on data retrieval technology, the real-time information of disasters and accidents is visualized and analyzed, which provide data support for emergency rescue and command decision-making. Finally, the above dual driving effects are combined to achieve accurate monitoring and reliable prediction of various mine disaster risks. The research results provide important technical support for improving the intelligent monitoring, safety assessment and emergency decision-making capabilities of mine disaster risks.

The article proposes a community embedded service facility health monitoring method that integrates Attention based Multimodal Temporal Graph Convolutional Network (AMT-GCN). Real time collection of water, electricity, and gas energy consumption and environmental parameters through Internet of Things architecture, design an attention mechanism driven cross modal feature fusion module, and construct an anomaly scoring model by combining spatiotemporal graph convolutional layers. The simulation results show that this method achieves an F1 of 0.92 in facility anomaly detection, and can achieve closed-loop governance of perception decision disposal, providing a reusable health management paradigm for embedded service facilities in urban communities.

In order to ensure the safety and stability of the home environment, a smart home safety monitoring system based on Internet of Things technology is designed. First, build a hierarchical system framework. Secondly, the software design is completed, the mixed encryption scheme is adopted to ensure the security of data transmission, and the intrusion detection mechanism is designed to prevent network intrusion. Finally, with the help of real-time monitoring and abnormal alarm strategy, potential safety hazards can be found in time, and convenient control of household equipment can be realized by remote control technology, and a three-level authority management system is designed to ensure control safety. The experimental results show that the system can realize the safe transmission and processing of home environment data, accurately identify network intrusion and environmental anomalies, and improve the security protection ability and management convenience of smart home.

In response to the challenges posed by the heterogeneous coexistence of 4G and 5G protocols on the integration capabilities of the baseband system, the scheduling mechanism, detection algorithm, resource mapping and power consumption control strategy of the single-chip dual-mode bearer architecture were studied. The design methods of collaborative scheduling modeling and asynchronous detection process were expounded, the dynamic resource reallocation and estimation structure optimization scheme were introduced, and experimental verification was conducted on a programmable platform. The experimental results show that the constructed algorithm system can effectively improve the system throughput, delay stability and energy efficiency level.

For small or medium-sized objects, in scenarios where high registration accuracy is required and processing time is limited, a strategy of combining two registration algorithms is adopted to balance registration accuracy and operational efficiency. A point cloud registration method based on the combination of Fast Global Registration (FGR) and Iterative Closest Point (ICP) is proposed. Point cloud preprocessing is accomplished through voxel downsampling and normal estimation; coarse registration is achieved using the fast point feature histogram in conjunction with FGR, obtaining an initial result, and feature extraction is performed only once to enhance efficiency; the point-to-surface point cloud fine registration algorithm is used to complete the fine registration. Experimental results show that the proposed method outperforms the other three comparison algorithms in terms of registration accuracy, and the processing time is significantly reduced, with stability.

To meet the demand for intelligent transformation of port machinery operations, an intelligent control system for portal cranes is constructed that integrates path optimization, state perception, and remote interaction. Design a remote control architecture and control strategy around the three functional modules of lifting trajectory planning, operating condition monitoring, and operation interface response, deploy a test platform for practical verification, and improve system operation efficiency and safety.

This study proposes a multi-sensor fusion-based Internet of things security monitoring system, which establishes a three-tier processing framework for data acquisition, preprocessing, and in-depth analysis through a hierarchical architecture. Field deployment verification demonstrates that the solution can keep end-to-end response latency below 800 ms while achieving a threat detection rate of 92.7% with only 0.1% false positive rate. The research findings provide innovative technical approaches for building highly reliable Internet of things security systems, offering significant application value in fields such as intelligent manufacturing.

Aiming at the problems of complex working conditions and inconsistent distribution of health status data faced by industrial equipment in actual operation, a prediction and optimization model of Remaining Useful Life (RUL) based on transfer learning is proposed. This model takes Deep Convolutional Neural Networks (DCNN) as the baseline model, integrates the feature extraction capabilities of Convolutional Neural Networks (CNN) and Long Short Term Memory (LSTM) networks, and combines the domain adaptive neural network method to realize the feature distribution alignment between the source domain and the target domain. Experiments show that the proposed model is significantly better than the benchmark model.

Aiming at the severe heat dissipation challenge faced by high-density Micro Light Emitting Diode (MicroLED) Augmented Reality (AR) optical machine in a limited space, a multi-dimensional collaborative heat dissipation design method of "material-structure-algorithm" is proposed. Magnesium alloy AZ91D is used to replace the traditional aluminum alloy bracket, and a graphene composite interface layer with three-dimensional Interpenetrating Network (IPN) is introduced on the back of the MicroLED chip. At the same time, the thermal path topology is optimized, and the IPN thermal conductive film is attached to the back of the MicroLED and led out from the cavity to the outside along the Flexible Printed Circuit (FPC) flat cable, which significantly improves the thermal diffusion efficiency; A dynamic temperature control model based on Back Propagation (BP) neural network is constructed. The ambient temperature, power consumption and time step are input to predict the temperature distribution and adjust the driving current in real time. The steady-state thermal analysis based on ANSYS Workbench shows that the collaborative design scheme reduces the peak temperature of the system from the initial 141.85℃ to below 50℃, with a decrease of 64.8%, and at the same time improves the uniformity of temperature distribution, which effectively solves the heat dissipation problem of high-integration MicroLED AR optical machine and provides feasible schemes and design theoretical basis for thermal management.

Based on the application requirements of semantic matching in agent tool call, a call framework for large-scale tool set is constructed. System design and implementation of query coding, tool modeling, approximate recall, collaborative call and continuous optimization mechanism. By demonstrating the core functions and cooperative relationship of each module, the experimental setup and performance evaluation methods in different scale environments are introduced. The research results show that the framework has obvious advantages in matching accuracy, call efficiency and system stability, and is suitable for efficient tool scheduling in complex scenarios.

Based on the challenges faced by current urban traffic management, a real-time monitoring system for urban traffic based on the Internet of Things was studied. The system architecture design, core function modules, data acquisition and processing mechanism and other key technologies are described, and the specific implementation methods of sensor node deployment, edge computing and real-time data transmission are introduced. Based on actual deployment, the performance and optimization strategies of the system were analyzed. The results indicate that the system can effectively improve the accuracy of traffic incident recognition and achieve rapid response, with strong practical application value.
Keywords: urban traffic monitoring; Internet of Things; multi-source data fusion