Offshore drilling platforms (ODPs) are critical infrastructure for exploring and developing marine oil and gas resources. As these platforms' capabilities expand, deploying intelligent surveillance services to ensure safe production has become increasingly important. However, the unique geographical locations and harsh environmental conditions of ODPs pose significant challenges for processing large volumes of video data, complicating the implementation of efficient surveillance systems. This study proposes a Cloud-Edge Large-Tiny Model Collaborative (CELTC) architecture grounded in deep reinforcement learning to optimize the processing and decision-making of surveillance data in offshore drilling platform scenarios. CELTC architecture leverages edge-cloud computing, deploying complex, high-precision large models on cloud servers and lightweight tiny models on edge devices. This dual deployment strategy capitalizes on tiny models' rapid response and large cloud models' high-precision capabilities. Additionally, the architecture integrates a deep reinforcement learning algorithm designed to optimize the scheduling and offloading of computational tasks between large and tiny models in the cloud-edge environment. The efficacy of the proposed architecture is validated using real-world surveillance data from ODPs through simulations and comparative experiments.
Unmanned aerial vehicles (UAVs) are a key driver of the low-altitude economy, where precise localization is critical for autonomous flight and complex task execution. However, conventional global positioning system (GPS) methods suffer from signal instability and degraded accuracy in dense urban areas. This paper proposes a lightweight and fine-grained visual UAV localization algorithm (FIM-JFF) suitable for complex electromagnetic environments. FIM-JFF integrates both shallow and global image features to leverage contextual information from satellite and UAV imagery. Specifically, a local feature extraction module (LFE) is designed to capture rotation, scale, and illumination-invariant features. Additionally, an environment-adaptive lightweight network (EnvNet-Lite) is developed to extract global semantic features while adapting to lighting, texture, and contrast variations. Finally, UAV geolocation is determined by matching feature points and their spatial distributions across multi-source images. To validate the proposed method, a real-world dataset UAVs-1100 was constructed in complex urban electromagnetic environments. The experimental results demonstrate that FIM-JFF achieves an average localization error of 4.03 m with a processing time of 2.89 s, outperforming state-of-the-art methods by improving localization accuracy by 14.9% while reducing processing time by 0.76 s.
As wireless communication technology advances, the convergence of the Internet of Vehicles (IoV) and edge computing is increasingly underpinning the progression of intelligent driving systems. Despite these advances, challenges persist in the stability and efficiency of vehicle-edge collaborative computing frameworks. This study introduces a Service Availability and Task Complexity-Aware Scheduling Framework for Intelligent Driving Tasks (SATCAS-IDT), ensuring the reliability of task scheduling between user vehicles and In-Vehicle Computing Platforms (IVCP) during high-speed mobility. Initially, a task complexity quantification model for intelligent driving is constructed, incorporating metrics such as information entropy, image contrast, energy, and color space. Subsequently, a vehicle mobility model tailored for high-speed multi-lane environments is designed, upon which the Mobility and Service Availability Aware IVCP Initial Screening Algorithm (MSAIF-IVCP) is developed, pre-selecting IVCPs that meet mobility and service availability requirements. Lastly, the Double Deep Recurrent Reinforcement Learning Task Scheduling Algorithm with Multi-observation Hybrid Encoder (DDRRL-MOHE) is proposed. By intelligently balancing task delay and processing accuracy, this algorithm dynamically adapts edge computing resources, effectively reducing response times and enhancing overall system performance. Experimental validation shows that our proposed method improves the task processing accuracy by about 1.70% on average and reduces the average latency by about 1.97% over the baseline method.
18 December 2025 in IEEE Transactions on Consumer Electronics
Panoramic cameras, offering a 360-degree field of view, greatly enhance the capabilities of consumer electronics, especially in applications like autonomous vehicles, where broad visibility is crucial for improving driving safety. However, this vast field of view introduces significant challenges for image processing tasks. Traditional visual processing models, reliant on customized solutions and specialized datasets, often increase development costs and resource inefficiencies. This study presents the Distributed Vehicular Panoramic Intelligent Service Architecture (DVPISA), leveraging advancements in 5G and edge computing. This framework employs pre-trained models, reducing the need for extensive data training and simplifying deployment. By incorporating techniques such as image splitting, task decomposition, and bipartite graph matching, DVPISA significantly enhances the speed and accuracy of panoramic image target detection. Empirical results show a mean Average Precision (mAP@0.5) of 78%, demonstrating its potential to set a new benchmark in consumer-grade vehicular electronics.
Remote sensing image change detection identifies surface changes by comparing images acquired at different time phases, offering crucial support for applications such as ecological monitoring, urban planning, disaster assessment, and resource management. However, the high resolution, complex noise, and diverse backgrounds of remote sensing images pose significant challenges, particularly in capturing subtle changes and mitigating noise interference. Existing methods often struggle with long-range dependency modeling, computational efficiency, and integrating shallow features and multiscale information, leading to suboptimal performance in detecting blurred boundaries and fine-grained changes. We propose the channel space fusion multiscale visual Mamba (CSFM-VMamba), an advanced change detection framework based on the visual Mamba architecture. CSFM-VMamba designs the channel space fusion block (CSFBlock) to jointly model channel and spatial attention, enabling flexible feature weight adjustments for more precise responses to changes. In addition, the multiscale selective fusion block (MSFBlock) facilitates dynamic fusion of multiscale features, enhancing sensitivity and robustness in detecting fine-grained changes. Comparative experiments on the SYSU-CD, LEVIR-CD+, WHU-CD, and SECOND datasets demonstrate that CSFM-VMamba consistently outperforms state-of-the-art methods, achieving performance improvements of 0.41% to 5.15% across various evaluation metrics. Ablation studies further validate the unique contributions of CSFBlock and MSFBlock, showcasing the potential of CSFM-VMamba in complex semantic change detection scenarios.