Unmanned aerial vehicles (UAV) play a vital role in many fields, such as agricultural planting, security patrol, emergency rescue, and so on. The development and implementation of these devices have become vital in terms of reachability and usability. Unfortunately, as drones become more widely used in various fields, they become more and more vulnerable to attacks and security threats, including, but not limited to, eavesdropping, man-in-the-middle attacks, and known session key attacks. In order to deal with these attacks and security threats and meet the needs of lightweight UAV communication, a secure and efficient authentication scheme is essential. To meet the security and lightweight requirements of an identity authentication scheme in a UAV communication network, this paper proposes an identity authentication scheme sdronelig based on an elliptic curve cryptosystem. The scheme realizes the mutual authentication and session key agreement configuration between the UAV and the ground station, and the authentication and key agreement between the UAVs can be realized with the help of the control station. The sdronelig authentication scheme is based on the ECDH key exchange protocol in the elliptic curve cryptography algorithm and adopts the MAC message authentication code technology and the method of pre-calculating part of the process. Under the premise of ensuring the security of the UAV communication network, the authentication efficiency is improved, the communication overhead and communication times are reduced, and the lightweight requirement of the UAV authentication scheme is met. Additionally, a formal verification tool is used to verify the security of the sdronelig scheme under the Dolev-Yao threat model, which is suitable for UAV networks. Finally, a detailed comparative study was conducted on security features, communication overhead, the number of communications, and computational overhead. The results show that the proposed sdronelig authentication scheme not only provides superior security features but also has better or comparable overhead compared to other existing authentication schemes.

Few-shot object detection (FSOD) in high-resolution remote sensing (RS) imagery remains challenging due to scarce annotations, large intra-class variability, and high visual similarity between categories, which together limit the generalization ability of convolutional neural network (CNN)-based detectors. To address this issue, we explore leveraging large vision-language models (LVLMs) for FSOD in RS. We propose a two-stage, parameter-efficient fine-tuning framework with hierarchical prompting that adapts Qwen3-VL for object detection. In the first stage, low-rank adaptation (LoRA) modules are inserted into the vision and text encoders and trained jointly with a Detection Transformer (DETR)-style detection head on fully annotated base classes under three-level hierarchical prompts. In the second stage, the vision LoRA parameters are frozen, the text encoder is updated using $K$-shot novel-class samples, and the detection head is partially frozen, with selected components refined using the same three-level hierarchical prompting scheme. To preserve base-class performance and reduce class confusion, we further introduce knowledge distillation and semantic consistency losses. Experiments on the DIOR and NWPU VHR-10.v2 datasets show that the proposed method consistently improves novel-class performance while maintaining competitive base-class accuracy and surpasses existing baselines, demonstrating the effectiveness of integrating hierarchical semantic reasoning into LVLM-based FSOD for RS imagery.

Road network information is a core component of online maps and plays a crucial role in navigation, urban planning, and traffic management. Convolutional Neural Networks (CNNs) have demonstrated remarkable performance in road extraction tasks. However, their limited ability to capture global information often leads to fragmented road segments when roads are occluded by other terrains in satellite images, ultimately undermining the accuracy and continuity of the segmentation results. Given the strengths of Transformers in capturing global contextual information and CNNs in extracting local detailed features, this letter introduces a novel deep network called Hybrid Segmentation Neural Network (HSNNet), which seamlessly integrates Transformers with CNNs to leverage the advantages of both architectures. To further enhance road continuity, we propose the road continuity perception module(RCPM). Experiments on the DeepGlobe and CHN6-CUG datasets demonstrate that our HSN-Net achieves state-of-the-art segmentation performance in road extraction tasks, validating the effectiveness of our design choices. The source code is available at https://github.com/hb281/HSN-Net.

Entity linking, a crucial task in the realm of natural language processing, aims to link entity mentions in a text to their corresponding entities in the knowledge base. While long documents provide abundant contextual information, facilitating feature extraction for entity identification and disambiguation, entity linking in Chinese short texts presents significant challenges. This study introduces an innovative approach to entity linking within Chinese short texts, combining multiple embedding representations. It integrates embedding representations from both entities and relations in the knowledge graph triples, as well as embedding representations from the descriptive text of entities and relations, to enhance the performance of entity linking. The method also incorporates external semantic supplements to strengthen the model’s feature learning capabilities. The Multi-Embedding Representation–Bidirectional Encoder Representation from Transformers–Bidirectional Gated Recurrent Unit (MER-BERT-BiGRU) neural network model is employed for embedding learning. The precision, recall, and F1 scores reached 89.73%, 92.18%, and 90.94% respectively, demonstrating the effectiveness of our approach.

In recent years, large-scale pre-trained models have developed rapidly in general fields, but there are few practically viable technical solutions for application scenarios in specialized fields, especially in sensitive areas such as healthcare, finance, and even military domains that require precise, professional, and credible knowledge for decision-making assistance. Concentrating data on cloud servers and then efficiently fine-tuning large-scale pre-trained models is impractical for industries with high data privacy protection requirements. This paper proposes a distributed training strategy for large-scale pre-trained models that integrates federated learning strategies. First, local data is fine-tuned offline on various edge devices. Then, the parameters of the efficiently fine-tuned small-scale models (e.g., LORA) are uploaded for aggregation on the server. Finally, intelligent services and solutions are provided for complex scenarios in specialized fields.