Traffic signs are pivotal components of traffic management, ensuring the regulation and safety of road traffic. However, existing detection methods often suffer from low accuracy and poor real-time performance in dynamic road environments. This paper reviews traditional traffic sign detection methods and introduces an enhanced detection algorithm (YOLO-BS) based on YOLOv8 (You Only Look Once version 8). This algorithm addresses the challenges of complex backgrounds and small-sized detection targets in traffic sign images. A small object detection layer was incorporated into the YOLOv8 framework to enrich feature extraction. Additionally, a bidirectional feature pyramid network (BiFPN) was integrated into the detection framework to enhance the handling of multi-scale objects and improve the performance in detecting small objects. Experiments were conducted on the TT100K dataset to evaluate key metrics such as model size, recall, mean average precision (mAP), and frames per second (FPS), demonstrating that YOLO-BS surpasses current mainstream models with mAP50 of 90.1% and FPS of 78. Future work will refine YOLO-BS to explore broader applications within intelligent transportation systems.

To increase the utilization rate of charging facilities and promote electric vehicles (EV), charging infrastructure should be built in a rational manner. The deployment strategy of public fast-charging stations (CS) determines whether charging demand can be met efficiently. Early CS planning has achieved some success. In the past, little is known about the charging behaviors and preferences of EV users, leading to a lack of relevance in the decisions made for the government-led CS construction. To address these issues, we used data on charging orders collected from public fast-CS s in cities for the summer and winter of 2023 to analyze charging user behavior patterns and build a more realistic model. Using queuing theory, we analyzed the facility size and operational metrics of public fast-CSs. Based on this, a capacity optimization model for public fast-CS was established to optimize the facility configuration of CSs by considering two factors: Investment cost and time cost. The optimal charging equipment configuration scheme for the public fast-CS is an 800 kVA charging box variable and 8 charging terminals, and the proposed facility scheme reduces the comprehensive cost by 28.1% and the user’s dwell time by 41.16%. If more consideration is given to the experience of EV users, the number of charging terminals can be set to 9.

Sandstorm weather significantly affects the quality of remote sensing images captured by drones, and collecting sandstorm datasets presents various challenges. This paper introduces a method for synthesizing a mild sandstorm dataset using Cycle-Consistent Adversarial Networks (CycleGAN), termed Sand-CycleGAN, which aimed to create a new sandstorm dataset, Sand-VisDrone, for the drone dataset VisDrone. This approach enhances the performance of remote sensing images in sandstorm conditions. To ensure that the synthesized dataset accurately reflects these conditions, we implement a Scatter Loss based on an atmospheric scatter model. This loss function improves the details and textures of the generated images, thereby increasing their realism. Additionally, we use a Super-Resolution Convolutional Neural Network (SRCNN) to enhance image resolution and restore detailed features that may be lost during the synthesis process. To validate the effectiveness of the Sand-VisDrone dataset, we conducted experiments with various dust removal networks and evaluated their performance using both reference-based and no-reference image quality metrics. The study also examines the influence of sandstorm image augmentation on the performance of object detection algorithms. Training results show that the mean Average Precision at IoU=0.5 (mAP@0.5) improved by 1.3% and 6.8% when using You Only Look Once version 10 narrow (YOLOv10n) and Real-Time Detection with Transformer for Object Tracking and Detection (RT-DETR) object detection algorithms, respectively.

Roadside unit (RSU) planning requires evidence about when vehicles may encounter candidate sites and whether a shortlist repeatedly covers the same vehicles. This study treats potential vehicle–site contact opportunities—not observed communication sessions or network traffic—as short-horizon vehicle-to-infrastructure (V2I) demand. Seven days of Hohhot taxi and bus trajectories, comprising more than 40 million records, were aggregated into 15 min panels for 1317 bus-stop-derived candidate sites. Training-only vehicle-transition relations augmented LightGBM forecasting, and a permuted-graph control assessed the value of the mobility structure. Historical vehicle-population overlap then informed set-level site screening, which was evaluated by future unique-vehicle coverage. Graph-informed forecasts showed modest improvements over tabular and permuted-graph models, while a historical time-of-day baseline remained stronger at 60 min. At the primary 100 m spatial footprint and 30 min horizon, overlap-aware screening improved unique-vehicle coverage by approximately 10% across shortlist budgets. Temporal-gap sensitivity and dependence-aware inference qualify the forecasting evidence. These results provide a data-driven screening basis for prioritizing candidate RSU sites before detailed feasibility, communications, and engineering assessment.

In the urban public transport network, the transfer of buses and subways provides convenience for residents to travel efficiently. But in actual operation, it is found that accidents, natural disasters, and other damage are inevitable. These sudden events may lead to route suspensions and service delays, ultimately resulting in network paralysis. In this paper, complex network theory is used to construct a weighted double-layer network model. Carrying capacity is considered the edge weight. The model analyzes the impact of these sudden events on network performance. It also conducts in-depth research on network structure and node importance. A collective influence (CI) algorithm is proposed as a centrality index to evaluate node importance. Based on the dynamic nature of the attacks, the network state is divided into initial network and current network. Taking Hohhot as an example, the results show that the network based on a CI algorithm node attack has the worst invulnerability. The network invulnerability based on an edge weight attack is better than that of edge betweenness. Compared with the current network, the invulnerability of the initial network is stronger. This indicates that ongoing changes and adaptations in the network may accelerate the decline in overall performance. At the same time, targeted interventions on key nodes and edges can enhance the network’s invulnerability. Planners can continuously monitor network performance to provide a basis for dynamic management and real-time adjustments. Additionally, effective information about critical routes to the public helps ensure the sustainable operation of the public transportation network.

Unmanned aerial vehicle (UAV) image transmission commonly relies on JPEG because of its interoperability and low deployment cost, yet compression can weaken structures used by downstream object detectors. This study presents JPEG-compatible receiver-side residual refinement (JCRR), a post-decoding pipeline that preserves the transmitter, JPEG syntax, and transmitted payload. JCRR comprises bounded residual reconstruction (BRR) and structure-prior residual refinement (SPRR), which adds a normalized Sobel edge prior. Both variants refine JPEG-decoded images before a fixed YOLO26s detector. An additional robustness check was conducted with a separately trained YOLO26n detector without retraining the receiver-side refinement models. Experiments used 548 VisDrone validation images at JPEG Q15, Q20, and Q25. BRR and SPRR were each trained with three seeds, and Artifact Reduction Convolutional Neural Network (ARCNN) was trained as an external post-decoding baseline under the same Q20-pair protocol. At the same transmitted JPEG payload, BRR increased mAP50 by 1.67–2.14 percentage points and SPRR by 1.67–2.22 percentage points. Under strict size bins defined after 640 × 640 letterboxing, SPRR improved small-target matched-target recall by 0.85–1.18 percentage points over JPEG; seed-specific paired bootstrap intervals were positive at every operating point. On the 300-image Q20 reconstruction holdout, SPRR achieved 37.75 ± 0.02 dB PSNR and 0.9779 ± 0.0002 SSIM, compared with 37.18 dB and 0.9763 for JPEG. Under the FP32 receiver-side forward-only timing protocol, BRR and SPRR processed images at approximately 62.0 and 61.6 frames per second, respectively, excluding JPEG decoding, detector inference, communication, and file I/O. WebP remained competitive at nearby, rather than bitrate-matched, operating points. The results support JCRR as a compatibility-oriented receiver-side enhancement for UAV systems that must retain JPEG transmission, while not implying bitrate reduction or end-to-end real-time flight deployment.