Multi-view scene matching refers to the establishment of a mapping relationship between images captured from different perspectives, such as those taken by unmanned aerial vehicles (UAVs) and satellites. This technology is crucial for the geo-localization of UAV views. However, the geometric discrepancies between images from different perspectives, combined with the inherent computational constraints of UAVs, present significant challenges for matching UAV and satellite images. Additionally, the imbalance of positive and negative samples between drone and satellite images during model training can lead to instability. To address these challenges, this study proposes a novel and efficient cross-view geo-localization framework called MSM-Transformer. The framework employs the Dual Attention Vision Transformer (DaViT) as the core architecture for feature extraction, which significantly enhances the modeling capacity for global features and the contextual relevance of adjacent regions. The weight-sharing mechanism in MSM-Transformer effectively reduces model complexity, making it highly suitable for deployment on embedded devices such as UAVs and satellites. Furthermore, the framework introduces a contrastive learning-based Symmetric Decoupled Contrastive Learning (DCL) loss function, which effectively mitigates the issue of sample imbalance between satellite and UAV images. Experimental validation on the University-1652 dataset demonstrates that MSM-Transformer achieves outstanding performance, delivering optimal matching results with a minimal number of parameters.

As the level of intelligence in agricultural consumer electronics continues to advance, data-driven devices are often faced with challenges such as resource shortages, high real-time requirements, and complex data processing. In particular, the diversity and variability of massive agricultural consumer electronics in the agricultural Internet of Things (IoT) complicate the resource allocation process. Traditional centralized cloud computing architectures often struggle to meet these demands. To address these issues, this paper proposes a resource collaborative allocation method for agricultural consumer electronics based on Service Function Chain (SFC). First, we formalize the resource requirements of various agricultural consumer electronics applications based on SFC in a resource mapping model. Subsequently, we develop a multidimensional resource balancing model with computing resources, storage resources, and bandwidth resources requirements of agricultural consumer electronics. We also devise a fine-grained end-to-end latency model to comprehensively consider the effective utilization of multidimensional resources and latency guarantees for agricultural consumer electronics. Finally, we employ a deep reinforcement learning algorithm with experience replay to optimize the resource allocation strategy in real-time, thereby enhancing the agricultural consumer electronics system’s responsiveness and resource utilization efficiency. Experimental results demonstrate that our proposed strategy outperforms existing methods in terms of improving resource utilization and reducing latency.

The increasing demand for accurate and accessible cancer detection has driven the development of AI-driven diagnostic systems on consumer-grade medical electronics. However, deploying large-scale deep learning models on these devices presents challenges due to limited computational resources, network constraints, and data privacy concerns. In this study, we propose an innovative Split Federated Learning (SFL) framework that synergistically combines the strengths of federated learning and split learning. The SFL framework introduces a novel hybrid architecture where deep learning models are partitioned between client-side and server-side components, enabling efficient local feature extraction while preserving data privacy through secure, encrypted intermediate feature transmission. Experiments conducted on the BreakHis dataset demonstrate the framework’s effectiveness, with the proposed method achieving a 93.1% accuracy while significantly reducing communication overhead. We further analyze the impact of dimensionality reduction techniques and model partitioning configurations to balance accuracy with efficiency. Results show that the SFL framework provides a practical, scalable solution for distributed cancer detection in real-world medical applications, offering a promising approach for privacy-preserving and bandwidth-efficient AI deployment on medical consumer electronics.

Connected and Autonomous Vehicles (CAVs) are pivotal for enabling intelligent mobility and autonomous driving, yet their inherent connectivity exposes critical vulnerabilities to cyber-attacks, which can precipitate life-threatening accidents, privacy breaches, and systemic failures in transportation networks. Given the escalating deployment of CAVs, ensuring robust cybersecurity is not merely a technical challenge but an urgent safety imperative. Traditional machine learning algorithms, such as Decision Trees (DT) and Support Vector Machines (SVM), often suffer from inadequate feature extraction in high-dimensional network traffic data, significantly compromising the accuracy of cyber-attack identification. Deep learning has emerged as a mainstream solution for Intrusion Detection Systems (IDS) due to its proficiency in handling complex data and automating feature extraction. To address these challenges, we propose LSTM-1DResNet, a novel deep learning-based intrusion detection model comprising an autoencoder and a classifier. The autoencoder innovatively integrates a Long Short-Term Memory Network (LSTM) with a one-dimensional convolutional residual module (Conv1D ResNet), substantially enhancing spatiotemporal feature extraction capabilities for high-dimensional traffic data. The classifier employs a Multilayer Perceptron (MLP) to deliver precise attack classification. The model was evaluated on both the NSL-KDD and CICIDS2017 datasets. On NSL-KDD, LSTM-1DResNet achieved 94.38% accuracy, outperforming standalone CNN and LSTM models by 11%. On CICIDS2017, it achieved 97.98% accuracy, precision of 98.11%, and recall of 97.98%, alongside high F1-scores(96.90%). This dual-dataset validation demonstrates the model's strong potential for enhancing intrusion detection in CAV-related contexts, particularly in addressing high-dimensional feature extraction challenges.

Steel plate fault detection is a knotty task in the manufacturing industry. Existing methods have achieved satisfactory performance, however, they still face challenges in accuracy, generalization ability, and capturing important features. Therefore, this study designs a steel plate fault detection method based on an Improved Set transformer and Kolmogorov-Arnold Network (ISet transformer-KAN). Firstly, a feature importance evaluation mechanism is incorporated into the Set transformer, dynamically optimizing seed vectors according to input data to better capture fault-related features. This modification improves information aggregation while leveraging the Set transformer’s inherent permutation invariance, ensuring stability regardless of feature order. Furthermore, KAN employs learnable activation functions with grid expansion mechanisms to progressively capture subtle feature details in steel plate faults, improving generalization capability and detection accuracy. Finally, to further validate the effectiveness of the proposed method, comprehensive experiments were conducted using the Steel Plate Faults dataset from the UCI Machine Learning Repository. The experimental results demonstrate that the proposed method outperforms traditional machine learning methods by an average of 10%-15% in accuracy and modern deep learning approaches by an average of 2%-3% in accuracy. The implementation code supporting the findings of this study is available at https://github.com/123fggv/maincode.