In Unmanned Aerial Vehicle (UAV) aerial photography, targets are usually small targets with dense distribution and unobvious features, and the object scale varies greatly. Therefore, the problems of missing detection and false detection are easy to occur in object detection. In order to solve these problems, a lightweight small object detection algorithm based on improved YOLOv8n, namely PECS-YOLO, is proposed for aerial photography. By adding P2 small object detection layer in the Neck part, the algorithm combines shallow and deep feature maps to better capture details of small targets. A lightweight convolution, namely PartialConv, is introduced to a new structure of Cross Stage Partial PartialConv (CSPPC), to replace Concatenation with Fusion (C2f) in the Neck network to realized lightweight of the model. By using a model of Spatial Pyramid Pooling with Efficient Layer Aggregation Network (SPPELAN), small object features can be captured effectively. By adding Squeeze-and-Excitation (SE)attention mechanism in front of each detection head in the Neck part, the network can better focus on useful channels and reduce the interference of background noise on small object detection tasks in complex environments. Finally, EfficiCIoU is used as the boundary frame loss function, and the shape difference of the boundary frame is also taken into account, which enhances the detection ability of the model for small targets. Experimental results show that, compared YOLOv8n, the mean Average Precision at Intersection over Union (IoU) of 0.5 (mAP@0.5) and the mean Average Precision at IoU of 0.5∶0.95 (mAP@0.5∶0.95) of PECS-YOLO object detection algorithm on VisDrone2019-DET dataset are increased by 3.5% and 3.7% respectively, the number of parameters is reduced by about 25.7%, and detection speed is increased by about 65.2%. In summary, PECS-YOLO model is suitable for small object detection in UAV aerial photography.

Steel surface defect detection technology in industrial scenarios is hindered by low detection accuracy and slow convergence speed. To address these issues, this study presents an improved YOLOv8 algorithm, namely a YOLOv8n-MDC. First, a Multi-scale Cross-fusion Network (MCN) is added to the backbone network. Establishing closer connections between the feature layers promotes uniform information transmission and reduces semantic information loss during cross-layer feature fusion, thereby enhancing the ability of the model to perceive steel defects. Second, deformable convolution is introduced in the module to adaptively change the shape and position of the convolution kernel, enabling a more flexible capture of the edge features of irregular defects, reducing information loss, and improving detection accuracy. Finally, a Coordinate Attention (CA) mechanism is added to embed position information into channel attention, solving the problem of position information loss and enabling the model to perceive the position and morphological features of defects, thereby enhancing detection precision and stability. Experimental results on the NEU-DET dataset show that the YOLOv8n-MDC algorithm achieves mAP@0.5 of 81.0%, which is 4.2 percentage points higher than that of the original baseline network. The algorithm has a faster convergence speed and higher accuracy; therefore, it meets the requirements of practical industrial production.

In view of missed and false detection phenomena caused by numerous small target instances and occlusions among targets in drone images, this paper proposes a lightweight small target detection algorithm for Unmanned Aerial Vehicle (UAV) images based on an improved YOLOv8. The Triple Feature Encoder (TFE) and Scale Sequence Feature Fusion (SSFF) modules are introduced in the neck to enhance the ability of the network to extract features at different scales. Furthermore, a Small Object Detection Head (SMOH) is designed and fused with the improved neck feature extraction network, and an additional detection head is also introduced to reduce the loss of small target features and enhance the recognition ability of the network for small targets. Additionally, considering the defects of Complete Intersection over Union (CIoU), a regression loss function, Wise-Inner-MPDIoU, is proposed by combining Wise-IoU, Inner-IoU, and Minimum Point Distance based IoU (MPDIoU). Finally, to realize the lightweight application requirements of the algorithm in mobile and embedded systems, amplitude-based layer-adaptive sparse pruning is performed to further reduce the model size while ensuring model accuracy. Experimental results demonstrate that, compared to the original YOLOv8s model, the improved model proposed in this paper improves mAP@0.5 by 6.8 percentage points, while reducing the number of parameters, amount of computation, and model size by 76.4%, 17.1%, and 73.5%, respectively. The proposed model is lightweight, improves detection accuracy, and has strong practical significance.

This study proposes a Common and Differential Cross-Attention Module-Bird's-Eye View (CDCAM-BEV) algorithm that combines 4D millimeter-wave radar and vision fusion to improve target detection accuracy for pedestrian and vehicle target recognition and localization in autonomous driving scenarios. First, a radar cylinder network is designed to encode the 4D radar point cloud into a pseudo image and convert the monocular image into a Bird's-Eye View (BEV) feature through Orthogonal Feature Transformation (OFT). Second, based on the cross-attention mechanism, a Common Information Extraction Module (CICAM) and a Differential Information Extraction Module (DICAM) are used to fully explore the common and differential information between radar and images. Finally, a BEV feature fusion module is designed based on CICAM and DICAM to achieve feature-level fusion of image and radar information in the BEV space. Experiments are conducted on the VOD dataset, and the CDCAM-BEV algorithm is compared with five other 3D object detection algorithms. The experimental results show that CDCAM-BEV achieves better detection performance in multiple modes. In the 3D mode, the average detection accuracy of CDCAM-BEV is 3.65 percentage points higher than that of the second ranked Part-A²; in the BEV mode, it is 5.04 percentage points higher than that of the second ranked PointPillars; in the Average Directional Similarity (AOS) mode, it is 2.62 percentage points higher than that of the second ranked Part-A². These results show that CDCAM-BEV exhibits excellent performance in all modes, effectively fusing images and 4D radar point cloud features, which significantly improves the accuracy and reliability of object detection.

Deep learning-based object detection has significantly improved the detection of medium and large targets. However, when detecting small objects, traditional algorithms often face challenges such as missed detections and false positives owing to the inherent issues of small scale and complex backgrounds. Therefore, this study aims to enhance the accuracy of small object detection by improving the YOLOv8 model. First, the convolutional module in the backbone is replaced with the RFAConv module, which enhances the ability of the model to process complex images. Second, a Mixed Local Channel Attention (MLCA) mechanism is introduced in the neck part, allowing the model to fuse features from different layers more efficiently while maintaining computational efficiency. Third, the Detect head of YOLOv8 is replaced with the Detect_FASFF head to address the inconsistency between different feature scales and improve the ability of the model to detect small objects. Finally, the Complete Intersection over Union (CIoU) loss function is replaced with the Focaler-IoU loss function, enabling the model to focus more on small objects that are difficult to locate precisely. Experimental results show that the improved model increases mAP@0.5 by 4.8 percentage points and mAP@0.5:0.95 by 3.0 percentage points on the FloW-Img dataset, which is sparse in small objects. On the VisDrone2019 dataset which has a high density of small objects, mAP@0.5 increases by 5.9 percentage points and mAP@0.5:0.95 improves by 4.0 percentage points. In addition, generalization comparison experiments are conducted on the low-altitude dataset AU-AIR and the pedestrian-dense detection dataset WiderPerson. The optimized model significantly improves the accuracy of small object detection compared with the original model and expands its applicability.

In the context of ongoing advancements in educational informatization, constructing precise and efficient curriculum knowledge graphs has become key to promoting personalized education development. As a structured knowledge representation model, curriculum knowledge graphs reveal complex relations between curriculum content and learning objectives to optimize the allocation of educational resources, and tailoring personalized learning paths for learners. This survey presents a discussion around the techniques used to construct curriculum knowledge graphs, starting with an explanation of the basic concepts; intrinsic connections; and significant differences among general, educational, and curriculum knowledge graphs. It then delves into the key technologies used for building curriculum knowledge graphs, covering aspects such as curriculum ontology design, entity extraction, and relation extraction, and provides a detailed analysis and summary of their evolution, key features, and limitations. Furthermore, it explores the application value of curriculum knowledge graphs in scenarios such as learning resource recommendation, learner behavior profile and modeling, and multimodal curriculum knowledge graph construction. Finally, it focuses on the challenges in constructing curriculum knowledge graphs, such as data diversity and heterogeneity, difficulties in quality evaluation, and the lack of cross-curriculum integration, and provides future-oriented insights based on cutting-edge technologies such as deep learning and Large Language Models (LLMs).

Large Language Models (LLMs) have made significant progress in dialogue, reasoning, and knowledge retention. However, they still face challenges in terms of factual accuracy, knowledge updates, and a lack of high-quality domain datasets for handling knowledge-intensive tasks in the electricity sector. This study aims to address these challenges by introducing an improved Retrieval-Augmented Generation (RAG) strategy. This strategy combines hybrid retrieval with a fine-tuned generative model for efficient knowledge capturing and updating. The Metadata-driven RAG framework (Meta-RAG) is proposed for knowledge Question Answering (QA) tasks in the electricity domain. This includes data preparation, model fine-tuning, and reasoning retrieval stages. The data-preparation stage involves document conversion, metadata extraction and enhancement, and document parsing. These processes ensure efficient indexing and structured processing of power regulation documents. The Electricity Question Answering (EleQA) dataset, consisting of 19 560 QA pairs, is constructed specifically for this sector. The model fine-tuning stage uses multi-question generation, chain-of-thought prompting, and supervised instruction fine-tuning to optimize the reasoning abilities in specific tasks. The retrieval reasoning stage employs mixed encoding and re-ranking strategies, combining retrieval and generation modules to improve answer accuracy and relevance. Experiments validate the effectiveness of Meta-RAG. Compared to baseline models such as Self-RAG, Corrective-RAG, Adaptive-RAG, and RA-ISF, Meta-RAG shows higher answer accuracy and retrieval hit rates. Meta-RAG with the Qwen1.5-14B-Chat model achieves an overall accuracy of 0.804 3, surpassing the other methods. Ablation and document recall experiments indicate that document retrieval significantly impacts the framework performance, with a 0.292 8 drop in accuracy when the retrieval capability is lost.

Deep learning has driven the development of artificial intelligence, which is widely used in computer vision. It provides breakthroughs and remarkable results in complex tasks such as image recognition, object detection, object tracking, and face recognition, demonstrating its excellent recognition and prediction capabilities. However, vulnerabilities and loopholes in deep learning models have been gradually exposed. Deep learning techniques, represented by convolutional neural networks, are extremely sensitive to well-designed adversarial examples, which can easily affect the security and privacy of the models. This paper first summarizes the concept of adversarial attacks, reasons for generating adversarial examples, and related terms. It outlines several types of classical adversarial attack strategies in the digital and physical domains and analyzes their advantages and disadvantages. Second, it focuses on computer vision and summarizes the latest research in adversarial attacks during tasks such as object detection, face recognition, object tracking, monocular depth estimation, and optical flow estimation, from both the digital and physical domains, as well as the various datasets commonly used in the study. It also briefly introduces the current stage of adversarial example defense and detection methods, summarizes the advantages and disadvantages of these methods, and describes examples of the applications of adversarial sample defense for various visual tasks. Finally, based on the summary of adversarial attack methods, it explores and analyzes the deficiencies and challenges of existing computer vision adversarial attacks.

Traditional machine learning algorithms perform well only when the training and testing sets are identically distributed. They cannot perform incremental learning for new categories or tasks that were not present in the original training set. Continual learning enables models to learn new knowledge adaptively while preventing the forgetting of old tasks. However, they still face challenges related to computation, storage overhead, and performance stability. Recent advances in pre-training models have provided new research directions for continual learning, which are promising for further performance improvements. This survey summarizes existing pre-training-based continual learning methods. According to the anti-forgetting mechanism, they are categorized into five types: methods based on prompt pools, methods with slow parameter updating, methods based on backbone branch extension, methods based on parameter regularization, and methods based on classifier design. Additionally, these methods are classified according to the number of phases, fine-tuning approaches, and use of language modalities. Subsequently, the overall challenges of continual learning methods are analyzed, and the applicable scenarios and limitations of various continual learning methods are summarized. The main characteristics and advantages of each method are also outlined. Comprehensive experiments are conducted on multiple benchmarks, followed by in-depth discussions on the performance gaps among the different methods. Finally, the survey discusses research trends in pre-training-based continual learning methods.

Pedestrian detection in crowded scenes is a key technology in intelligent monitoring of public space. It enables the intelligent monitoring of crowds, using object detection methods to detect the positions and number of pedestrians in videos. This paper presents Crowd-YOLOv8, an improved version of the YOLOv8 detection model, to address the issue of pedestrians being easily missed owing to occlusion and small target size in densely populated areas. First, nostride-Conv-SPD is introduced into the backbone network to enhance its capability of extracting fine-grained information, such as small object features in images. Second, small object detection heads and the CARAFE upsampling operator are introduced into the neck part of the YOLOv8 network to fuse features at different scales and improve the detection performance in the case of small targets. Experimental results demonstrate that the proposed method achieves an mAP@0.5 of 84.3% and an mAP@0.5∶0.95 of 58.2% on a CrowdedHuman dataset, which is an improvement of 3.7 and 5.2 percentage points, respectively, compared to those of the original YOLOv8n. On the WiderPerson dataset, the proposed method achieves an mAP@0.5 of 88.4% and an mAP@0.5∶0.95 of 67.4%, which is an improvement of 1.1 and 1.5 percentage points compared to those of the original YOLOv8n.

Federated learning enables participants to collaboratively model without revealing their raw data, thereby effectively addressing the privacy issue of distributed data. However, as research advances, federated learning continues to face security concerns such as privacy inference attacks and malicious client poisoning attacks. Existing improvements to federated learning mainly focus on either privacy protection or against poisoning attacks without simultaneously addressing both types of attacks. To address both inference and poisoning attacks in federated learning, a privacy-preserving against poisoning federated learning scheme called APFL is proposed. This scheme involves the design of a model detection algorithm that utilizes Differential Privacy (DP) techniques to assign corresponding aggregation weights to each client based on the cosine similarity between the models. Homomorphic encryption techniques are employed for the weighted aggregation of the local models. Experimental evaluations of the MNIST and CIFAR10 datasets demonstrate that APFL effectively filters malicious models and defends against poisoning attacks while ensuring data privacy. When the poisoning ratio is no more than 50%, APFL achieves a model performance consistent with the Federated Averaging (FedAvg) scheme in a non-poisoned environment. Compared with the Krum and FLTrust schemes, APFL exhibits average reductions of 19% and 9% in model test error rate, respectively.

As multivariate time series data become increasingly prevalent across various industries, anomaly detection methods that can ensure the stable operation and security of systems have become crucial. Owing to the inherent complexity and dynamic nature of multivariate time series data, higher demands are placed on anomaly detection algorithms. To address the inefficiencies of existing anomaly detection methods in processing high-dimensional data with complex variable relations, this study proposes an anomaly detection algorithm for multivariate time series data, based on Graph Neural Networks (GNNs) and a diffusion model, named GRD. By leveraging node embedding and graph structure learning, GRD algorithm proficiently captures the relations between variables and refines features through a Gated Recurrent Unit (GRU) and a Denoising Diffusion Probabilistic Model (DDPM), thereby facilitating precise anomaly detection. Traditional assessment methods often employ a Point-Adjustment (PA) protocol that involves pre-scoring, substantially overestimating an algorithm's capability. To reflect model performance realistically, this work adopts a new evaluation protocol along with new metrics. The GRD algorithm demonstrates F1@k scores of 0.741 4, 0.801 7, and 0.767 1 on three public datasets. These results indicate that GRD algorithm consistently outperforms existing methods, with notable advantages in the processing of high-dimensional data, thereby underscoring its practicality and robustness in real-world anomaly detection applications.

Blockchain, as a distributed and trusted database, has gained significant attention in academic and industrial circles for its effective application in the domain of digital copyright protection. Traditional digital copyright protection technologies suffer from issues such as difficulties in tracking infringements, complexities in copyright transactions, and inadequate protection of legitimate rights, which severely hampering the development of digital copyright protection endeavors. The immutability, traceability, and decentralization inherent in blockchain technology provide a highly reliable, transparent, and secure solution to mitigate the risks associated with digital copyright infringement. This overview starts with an introduction to the fundamental principles of blockchain technology. Then, it discusses the latest research findings on the integration of blockchain with traditional copyright protection technologies to address the problems inherent in traditional copyright protection schemes. Further, an evaluation of the practical applications and potential of blockchain is conducted, emphasizing its positive impact on the copyright protection ecosystem. Finally, this overview delves into the challenges and future trends related to blockchain based copyright protection, ultimately aiming to establish a more robust and sustainable blockchain copyright protection system.

Artificial intelligence has made remarkable progress across many fields, encouraging countries to attach great importance to its research and development. However, the rapid development of artificial intelligence has also brought about a series of problems and threats, and overreliance on and blind trust in such models can lead to serious risks. Therefore, interpretable artificial intelligence has become a key element in building trusted and transparent intelligent systems, and its research and development requires immediate attention. This survey comprehensively summarizes the research progress on explainable artificial intelligence at home and abroad comprehensively from multiple dimensions and levels. Based on current research results in the industry, this survey subdivides the key technologies of explainable artificial intelligence into four categories: interpretation model, interpretation method, safety testing, and experimental verification, with the aim of clarifying the technical focus and development direction of each field. Furthermore, the survey explores specific application examples of explainable artificial intelligence across key industry sectors, including but not limited to education, healthcare, finance, autonomous driving, and justice, demonstrating the significant role it plays in enhancing decision-making transparency. Finally, this survey provides an in-depth analysis of the major technical challenges of interpretable artificial intelligence and presents future development trends, in addition to a special investigation and in-depth analysis of the interpretability of large models, which has attracted considerable attention recently.

The sequence recommendation algorithm dynamically models the user's historical behavior to predict the content they may be interested in. This study focuses on the application of contrastive Self Supervised Learning (SSL) in sequence recommendation, enhancing the model's representation ability in sparse data scenarios by designing effective self supervised signals. First, a personalized data augmentation method incorporating user preferences is proposed to address the issue of noise introduced by random data augmentation. This method guides the augmentation process based on user ratings and combines different augmentation methods for short and long sequences to generate augmented sequences that align with user preferences. Second, a mixed-augmentation training approach is designed to address the issue of imbalanced feature learning during training. In the early stages of training, augmentation sequences are generated using randomly selected methods to enhance the model performance and generalization. In the later stages, augmentation sequences with high similarity to the original sequences are selected to enable the model to comprehensively learn the actual preferences and behavior patterns of users. Finally, traditional sequence prediction objectives are combined with SSL objectives to infer user representations. Experimental verification is performed using the Beauty, Toys, and Sports datasets. Compared with the best result in the baseline model, the HR@5 indicator of the proposed method increases by 6.61%, 3.11%, and 3.76%, and the NDCG@5 indicator increases by 11.40%, 3.50%, and 2.16%, respectively, for the aforementioned datasets. These experimental results confirm the rationality and validity of the proposed method.

Intent recognition is important in natural language understanding. Previous research on intent recognition has primarily focused on single-modal intent recognition for specific tasks. However, in real-world scenarios, human intentions are complex and must be inferred by integrating information such as language, tone, expressions, and actions. Therefore, a novel attention-based multimodal fusion method is proposed to address intent recognition in real-world multimodal scenarios. To capture and integrate the long-range dependencies between different modalities, adaptively adjust the importance of information from each modality, and provide richer representations, a separate self-attention mechanism is used for each modality feature. By adding explicit modality identifiers to the data of each modality, the model can distinguish and effectively fuse information from different modalities, thereby enhancing overall understanding and decision-making capabilities. Given the importance of textual information in cross-modal interactions, a multimodal fusion method based on a cross-attention mechanism is employed, with text as the primary modality and other modalities assisting and guiding the interactions. This approach aims to facilitate interactions among textual, visual, and auditory modalities. Finally, experiments were conducted on the MIntRec and MIntRec2.0 benchmark datasets for multimodal intent recognition. The results show that the model outperforms existing multimodal learning methods in terms of accuracy, precision, recall, and F1 score, with an improvement of 0.1 to 0.5 percentage points over the current best baseline model.

To optimize Quality of Service (QoS), Mobile Edge Computing (MEC) has been deeply integrated into the Internet of Vehicle (IoV) to provide geographically proximal computing resources for vehicles, thereby reducing task processing latency and energy consumption. However, traditional MEC server deployment relies primarily on terrestrial Base Stations (BSs), resulting in high deployment costs and limited coverage, making it difficult to ensure uninterrupted services for all vehicles. Air-ground collaborative IoV technology has emerged as a solution to these challenges. Unmanned Aerial Vehicles (UAVs) can dynamically assist Road-Side Units (RSUs) using their flexibility in line-of-sight links, providing more flexible computing resources for vehicular users, thereby ensuring the continuity and efficiency of in-vehicle services. Therefore, this study proposes a Dynamic Vehicular Edge Task Offloading Method (DVETOM) based on air-ground collaboration. This method adopts a vehicle-road-air architecture, establishing Vehicle-to-RSU (V2R) and Vehicle-to-UAV (V2U) links. Transmission and computation models are constructed for three modes: local execution of vehicular tasks, offloading tasks to the RSU, and offloading tasks to the UAV. An objective function is established with the joint optimization goal of minimizing system latency and energy consumption. DVETOM transforms the task offloading problem into a Markov Decision Process (MDP) and optimizes the task offloading strategy by using the Distributed Deep Deterministic Policy Gradient (D4PG) algorithm based on Deep Reinforcement Learning (DRL). Compared with 5 benchmark methods, experimental results show that DVETOM outperforms existing methods by 3.45%—23.7% in terms of reducing system latency and 5.8%—23.47% in terms of reducing system energy consumption while improving QoS for vehicular users. In conclusion, DVETOM enhances the offloading of vehicular edge computing tasks within the IoV effectively. It offers IoV users a more efficient and energy-conserving solution, showcasing its extensive potential for application in intelligent transportation systems.

As the main window of human-computer interaction, the mobile phone screen has become an important factor affecting the user experience and the overall performance of the terminal. As a result, there is a growing demand to address defects in mobile phone screens. To meet this demand, in view of the low detection accuracy, high missed detection rate of small target defects, and slow detection speed in the process of defect detection on mobile phone screens, a PGS-YOLO algorithm is proposed, with YOLOv8n as the benchmark model. PGS-YOLO effectively improves the detection ability of small targets by adding a special small target detection head and combining it with the SeaAttention attention module. The backbone and feature fusion networks are integrated into PConv and GhostNetV2 lightweight modules, respectively, to ensure accuracy, reduce the number of model parameters, and improve the speed and efficiency of defect detection. The experimental results show that, in the dataset of mobile phone screen surface defects from Peking University, compared with the results of YOLOv8n, the mAP@0.5 and mAP@0.5∶0.95 of the PGS-YOLO algorithm are increased by 2.5 and 2.2 percentage points, respectively. The algorithm can accurately detect large defects in the process of mobile phone screen defect detection as well as maintain a certain degree of accuracy for small defects. In addition, the detection performance is better than that of most YOLO series algorithms, such as YOLOv5n and YOLOv8s. Simultaneously, the number of parameters is only 2.0×10⁶, which is smaller than that of YOLOv8n, meeting the needs of industrial scenarios for mobile phone screen defect detection.

Simulators play an indispensable role in an array of scientific fields involving research and development. Particularly in architectural design, simulators provide a secure and cost-effective virtual environment, enabling researchers to conduct rapid experimental analyses and evaluations. Simultaneously, simulators facilitate the acceleration of the chip design and verification processes, thereby conserving time and reducing resource expenditure. However, with the evolutionary advances in processor architectural designs—specifically, the flourishing diversifications featured in dedicated processors—the key role played by simulators in providing substantial feedback for architectural design exploration has gained prominence. This discourse provides an overview of the current developments and applications of architectural simulators, accentuating a few illustrative examples. Analyzing the techniques employed by simulators dedicated to various processors allows for a deeper understanding of the focal points and technical complexities under different architectures. Moreover, this discourse deliberates speculative assessments and critiques of vital aspects of future architectural simulator developments, aspiring to forecast their prospects in the field of processor design research.

In recent years, Large Language Models (LLMs) such as GPT, LLaMA, Qwen, and DeepSeek, have achieved significant breakthroughs in natural language processing, computer vision, multimodal learning, and other fields. However, constrained by factors such as their reasoning mechanisms, parameter scales, and the inherent knowledge contained within their training data, these models often suffer from issues like ″hallucinations″—characterized by inaccurate answers and even factual deviations—when handling complex tasks, addressing questions from professional domains, or generating time-sensitive content. These limitations severely hinder their application in high-reliability scenarios. The ″tool learning″ paradigm is attracting increasing attention as a promising solution to these capability bottlenecks. Its primary objective is to enable LLMs to understand and utilize external tools to complete specific tasks. By invoking external tools, such as databases, search engines, and mathematical tools, LLMs can transcend their parameterized knowledge; enhance their reasoning, decision-making, and execution capabilities; and mitigate hallucination problems. This paper systematically reviews the development context and technical advancements in LLM tool learning, analyzes the expansion of LLM capabilities through tools, summarizes tool invocation mechanisms ranging from in-context learning to fine-tuning training, and discusses key issues including performance optimization and adaptive tool generation. The paper also analyzes evaluation methods for LLM tool invocation, summarizes the current challenges in tool learning, and outlines future research directions.

The current high-precision vehicle detection model faces challenges due to its excessive parameterization and computational demands, making it unsuitable for efficient operation on intelligent transportation devices. Conversely, lightweight vehicle detection models often sacrifice accuracy, rendering them unsuitable for practical tasks. In response, an improved lightweight vehicle detection network based on YOLOv8 is proposed. This enhancement involves substituting the main network with the FasterNet architecture, which reduces the computational and memory access requirements. Additionally, we replace the Bidirectional Feature Pyramid Network (BiFPN) in the neck with a weighted bidirectional feature pyramid network to simplify the feature fusion process. Simultaneously, we introduce a dynamic detection head with a fusion attention mechanism to achieve nonredundant integration of the detection head and attention. Furthermore, we address the deficiencies of the Complete Intersection over Union (CIoU) in terms of detection accuracy and convergence speed by proposing a regression loss algorithm that incorporates the Scale-invariant Intersection over Union (SIoU) combined with the Normalized Gaussian Wasserstein Distance (NWD). Finally, to minimize the computational demands on edge devices, we implement amplitude-based layer-wise adaptive sparsity pruning, which further compresses the model size. Experimental results demonstrate that the proposed improved model, compared with the original YOLOv8s model, achieves a 1.5 percentage points increase in accuracy, a 78.9% reduction in parameter count, a 67.4% decrease in computational demands, and a 77.8% reduction in model size. This demonstrates the outstanding lightweight effectiveness and practical utility of the proposed model.

A multimodal remote sensing small-target detection method, BFMYOLO, is proposed to address misdetection and omission issues in remote sensing images with complex backgrounds and less effective information. The method utilizes a pixel-level Red-Green-Blue (RGB) and infrared (IR) image fusion module, namely, the Bimodal Fusion Module (BFM), for effectively making full use of the complementarity of different modes to realize the effective fusion of information from two modalities. In addition, a full-scale adaptive updating module, AA, is introduced to resolve multitarget information conflicts during feature fusion. This module incorporates the CARAFE up-sampling operator and shallow features to enhance non-neighboring layer fusion and improve the spatial information of small targets. An Improved task decoupling Detection Head (IDHead) is designed to handle classification and regression tasks separately, thereby reducing the mutual interference between different tasks and enhancing detection performance by fusing deeper semantic features. The proposed method adopts the Normalized Wasserstein Distance (NWD) loss function as the localization regression loss function to mitigate positional bias sensitivity. Results of experiments on the VEDAI, NWPU VHR-10, and DIOR datasets demonstrate the superior performance of the model, with mean Average Precision when the threshold is set to 0.5 (mAP@0.5) of 78.6%, 95.5%, and 73.3%, respectively. The model thus outperforms other advanced models in remote sensing small-target detection.

Convolutional Neural Networks (CNNs) are widely used in the field of object detection, earning widespread acclaim in scholarly circles due to their precision and scalability. It has spawned numerous notable models, including those in the Region-based Convolutional Neural Networks (R-CNNs) (such as Fast R-CNN and Faster R-CNN) and You Only Look Once (YOLO) series. After the success of Transformers in the field of natural language processing, researchers began exploring their application in computer vision, leading to the development of visual backbone networks such as Visual Transformer (ViT) and Swin Transformer. In 2020, a Facebook research team unveiled DEtection TRansformer (DETR), an end-to-end object detection algorithm based on Transformers, designed to minimize the need for prior knowledge and postprocessing in object detection tasks. Despite the promise shown by DETR in object detection, it has limitations including low convergence speed, relatively low accuracy, and the ambiguous physical significance of target queries. These issues have spurred a wave of research aimed at refining and enhancing the algorithm. This paper aims to collate, scrutinize, and synthesize the various efforts aimed at improving DETR, assessing their respective merits and demerits. Furthermore, it presents a comprehensive overview of state-of-the-art research and specialized application domains that employ DETR and concludes with a prospective analysis of the future role of DETR in the field of computer vision.

The accurate recognition of fake news is an important research topic in the online environment, where distinguishing information explosion and authenticity is difficult. Existing studies mostly use multiple deep learning models to extract multivariate semantic features to capture different levels of semantic information in the text; however, the simple splicing of these features causes information redundancy and noise, limiting detection accuracy and generalization, and effective deep fusion methods are not available. In addition, existing studies tend to ignore the impact of dual sentiments co-constructed by news content and its corresponding comments on revealing news authenticity. This paper proposes a Dual Emotion and Multi-feature Fusion based Fake News Detection (DEMF-FND) model to address these problems. First, the emotional features of news and comments are extracted by emotion analysis. The emotional difference features reflecting the correlation between the two are introduced using similarity computation, and a dual emotion feature set is constructed. Subsequently, a fusion mechanism based on multihead attention is used to deeply fuse the global and local semantic features of the news text captured by a Bidirectional Long Short-Term Memory (BiLSTM) network with a designed Integrated Static-Dynamic Embedded Convolutional Neural Network (ISDE-CNN). Eventually, the dual emotion feature set is concatenated with the semantic features obtained by deep fusion and fed into a classification layer consisting of a fully connected layer, to determine news authenticity. Experimental results show that the proposed method outperforms the baseline method in terms of benchmark metrics on three real datasets, namely Weibo20, Twitter15, and Twitter16, and achieves 2.5, 2.3, and 5.5 percentage points improvements in accuracy, respectively, highlighting the importance of dual emotion and the deep fusion of semantic features in enhancing the performance of fake news detection.

Acoustic Scene Classification (ASC) aims to enable computers to simulate the human auditory system in the task of recognizing various acoustic environments, which is a challenging task in the field of computer audition. With rapid advancements in intelligent audio processing technologies and neural network learning algorithms, a series of new algorithms and technologies for ASC have emerged in recent years. To comprehensively present the technological development trajectory and evolution in this field, this review systematically examines both early work and recent developments in ASC, providing a thorough overview of the field. This review first describes application scenarios and the challenges encountered in ASC and then details the mainstream frameworks in ASC, with a focus on the application of deep learning algorithms in this domain. Subsequently, it systematically summarizes frontier explorations, extension tasks, and publicly available datasets in ASC and finally discusses the prospects for future development trends in ASC.

As a mainstream tool for container orchestration, Kubernetes can support automatic deployment, service discovery, and load balancing. It is known for its high availability and performance. However, scheduling strategies such as the best adaptation algorithm or the minimum negative cut method ignore the heterogeneity and energy differences of nodes. In addition, Kubernetes tools only consider CPU and memory resources and set a unified weight mechanism in advance, which can easily lead to problems such as load imbalance, performance degradation, and the inability to satisfy refined scheduling. To address these existing problems, this study proposes a heterogeneous task scheduling algorithm based on multi-dimensional resources, namely A-KCSS. The A-KCSS algorithm is based on the heterogeneous computing resources of a cluster. It adds disk Input/Output (I/O), network I/O load, and GPU resources as evaluation indicators for filtering and screening, and it more comprehensively considers the heterogeneity of nodes. This study also introduces a weight calculation model based on multi-dimensional resource factors. Based on the resource requirements of the task to be scheduled, the weight value of each dimension of the resource factor of the task to be scheduled is calculated, and the score of each node is calculated based on the real-time resource utilization of the cluster. Nodes are prioritized according to the score, and the node with the highest priority is selected for scheduling. The performance of the A-KCSS algorithm is experimentally verified on the Kubernetes cluster. Compared with the default scheduling algorithm and the KCSS algorithm, the average response time is reduced by 10% and 4%, the throughput is increased by 30% and 15%, the availability is improved by 40% and 30%, and the load balancing performance is increased by 23% and 18%, respectively, thereby improving the overall cluster performance.

Existing object detection algorithms suffer from low detection accuracy and poor real-time performance when detecting fall events in indoor scenes, owing to changes in angle and light. In response to this challenge, this study proposes an improved fall detection algorithm based on YOLOv8, called OEF-YOLO. The C2f module in YOLOv8 is improved by using a Omni-dimensional Dynamic Convolution (ODConv) module, optimizing the four dimensions of the kernel space to enhance feature extraction capabilities and effectively reduce computational burden. Simultaneously, to capture finer grained features, the Efficient Multi-scale Attention (EMA) module is introduced into the neck network to further aggregate pixel-level features and improve the network's processing ability in fall scenes. Integrating the Focal Loss idea into the Complete Intersection over Union (CIoU) loss function allows the model to pay more attention to difficult-to-classify samples and optimize overall model performance. Experimental results show that compared to YOLOv8n, OEF-YOLO achieves improvements of 1.5 and 1.4 percentage points in terms of mAP@0.5 and mAP@0.5∶0.95, the parameters and computational complexity are 3.1×10⁶ and 6.5 GFLOPs. Frames Per Second (FPS) increases by 44 on a Graphic Processing Unit (GPU), achieving high-precision detection of fall events while also meeting deployment requirements in low computing scenarios.

The RSD-YOLO algorithm, based on YOLOv8s, is proposed to address the challenges of low detection performance, severe occlusion, difficulty of small target feature extraction, and large number of model parameters inherent in Unmanned Aerial Vehicle (UAV) aerial images. First, the Receptive Field Attention (RFA) module CSP-RFA is designed to replace the C2f module for enhancing the capability of small target feature extraction, effectively addressing the insensitivity of traditional convolutional operations to positional changes. Second, the backbone and feature fusion networks are made lightweight, a new large-size feature map detection branch is added, and a Receptive Field Pyramid Network (RFPN) is proposed to optimize the feature flow direction and improve feature representation. Third, the detection head module is optimized by integrating multi-scale features with a multi-level attention mechanism and the loss function is updated to improve the model’s detection performance for small targets. Finally, in terms of model compression, Layer-Adaptive Magnitude-based Pruning (LAMP) algorithm is employed to further reduce the number of parameters and model size. The experimental results demonstrate that the lightweight RSD-YOLO model significantly outperforms the baseline model on the publicly available VisDrone2019 dataset, with a 10.0 percentage point increase in precision, a 9.5 percentage point increase in mAP@0.5 (equivalent to a 24.1% increase), and a 6.9 percentage point increase in mAP@0.5∶0.95 (equivalent to a 29.4% increase). The number of model parameters is reduced from 11.12×10⁶ to 4.05×10⁶, representing a 63.6% reduction, and the computational cost is reduced from 42.7 GFLOPs to 25.5 GFLOPs, showing a 40% reduction. Furthermore, for a newly filtered dataset focusing on small occluded targets, RSD-YOLO shows improvements of 9.1, 16.1, and 10.7 percentage points in terms of precision, mAP@0.5, and mAP@0.5∶0.95, respectively.

The primary task of person Re-IDentification (ReID) is to identify and track a specific pedestrian across multiple non-overlapping cameras. With the development of deep neural networks and owing to the increasing demand for intelligent video surveillance, ReID has gradually attracted research attention. Most existing ReID methods primarily adopt labeled data for supervised training; however, the high annotation cost makes the scaling supervised ReID to large unlabeled datasets challenging. The paradigm of unsupervised ReID can significantly alleviate such issues. This can improve its applicability to real-life scenarios, enhancing its research potential. Although several ReID surveys have been published, they have primarily focused on supervised methods and their applications. This survey systematically reviews, analyzes, and summarizes existing ReID studies to provide a reference for researchers in this field. First, the ReID methods are comprehensively reviewed in an unsupervised setting. Based on the availability of source domain labels, the unsupervised ReID methods are categorized into unsupervised domain adaptation methods and fully unsupervised methods. Additionally, their merits and drawbacks are discussed. Subsequently, the benchmark datasets widely evaluated in ReID research are summarized, and the performance of different ReID methods on these datasets is compared. Finally, the current challenges in this field are discussed and potential future directions are proposed.

To solve the problems of poor detection effect, high misdetection and omission rate, and weak generalization ability of urban vehicle target detection algorithms, this study proposes an improved YOLOv8 urban vehicle target detection algorithm. First, an Efficient Multi-scale Attention (EMA) mechanism is incorporated into the tail of the backbone network, which helps the model better capture the detailed features of a target vehicle. Combined with a 160×160 pixel small-target detection layer, it enhances the detection capability of small targets and aggregates pixel-level features through dimensional interaction to enhance the mining capability of the target vehicle. Second, the study designs a new Multi-scale Lightweight Convolution (MLConv) module for the lightweight network, and the C2f module is reconstructed based on MLConv, which significantly improves the feature extraction capability of the model. Finally, to suppress the harmful gradients generated by low-quality images, the study uses the Wise-Intersection over Union (WIoU) loss function instead of the Complete Intersection over Union (CIoU) to optimize the network's bounding box loss and improve the model's convergence speed and regression accuracy. On the Streets vehicle dataset, the algorithm improves mAP@0.5, mAP@0.5∶0.95, and recall by 1.9, 1.4 and 2.4 percentage points respectively, compared with the YOLOv8n benchmark model. In validations on a domestic vehicle dataset and the VisDrone2019 small target dataset, these performance indexes improve to different degrees, proving that the improved algorithm has good generalization and robustness. Compared with other mainstream algorithms, the improved algorithm exhibits higher accuracy and detection rate, indicating that the algorithm performs better in urban vehicle target detection.

Distribution network planning is important in power systems because it directly affects the reliability, efficiency, and economy of power supply. Good planning ensures that power resources are allocated efficiently while reducing operating costs and power losses. However, as power demand and system complexity increase, traditional decision-making methods are no longer applicable. To improve the efficiency and reliability of equipment selection, connection configuration, and grid layout, this study proposes an intelligent distribution network planning method based on Knowledge Graphs (KGs) and Graph Convolutional Neural Networks (GCNNs), i.e., KG-GCNN. This method leverages the advantages of KG, Graph Neural Networks (GNNs) and Convolutional Neural Networks (CNNs). This study also provides an intelligent distribution network planning method for power system planners to better understand, analyze, and optimize the equipment configuration, connection, and physical layout of power systems. The study first establishes the KG of the power network, which includes the equipment, properties, and interrelationships, and provides the basis for subsequent analysis and optimization. Then, it uses a GNNs to analyze the structural data of the power network to capture the relationship and influence between the devices and to provide important information for equipment configuration and connection decisions. Finally, it introduces CNNs to improve the physical layout of the grid to determine the best location and connection for the devices in the grid, thereby improving its performance and reliability. The experimental results show that, compared with decision trees, Support Vector Machines (SVMs), and Recurrent Neural Networks (RNNs), the proposed method can effectively match the complex topologies of power grids and is suitable for optimizing the physical layout of power grids.

Traffic signal control plays an important role in alleviating traffic congestion and improving urban commuting efficiency. In recent years, breakthroughs have been made in traffic signal control algorithms based on deep reinforcement learning using real-time traffic data as input. However, traffic data in real-world scenarios often involve data distortion. Traditional solutions use reinforcement learning algorithms to control signal lights after repairing distorted data. However, on the one hand, the dynamic phases of traffic signal introduces additional uncertainty to distortion repair, and on the other hand, distortion repair is difficult to combine with deep reinforcement learning frameworks to improve performance. To address these issues, a distorted traffic signal control model based on hidden state prediction, HCRL, is proposed. The HCRL model comprises encoding, control, and encoding prediction sub-models. By introducing a hidden state representation mechanism for signalized intersections, the HCRL model can adapt better to deep reinforcement learning frameworks and effectively express the control state of signalized intersections. In addition, the HCRL model uses a special transfer training method to avoid data distortion interference in the control sub-model. Two real datasets are used to verify the impact of data distortion on the intelligent signal light control algorithms. The experimental results show that the HCRL model outperforms the distortion-completion-based traffic signal control models in all distortion scenarios and distortion rates; further, it demonstrates strong robustness against data distortion when compared with other baseline models.

The Siamese tracking network is a popular target tracking framework that includes three modules: backbone, fusion, and positioning networks. The Transformer is a relatively new and effective implementation method for fusion network modules. The encoder and decoder of the Transformer use a self-attention mechanism to enhance the features of the Convolutional Neural Network (CNN). However, the self-attention mechanism can only enhance features in the spatial dimension without considering feature enhancement in the channel dimension. To enable the self-attention network of the Transformer to enhance features both in the spatial and channel dimensions and provide accurate correlation information for the target localization network, a Transformer tracker based on dual-dimensional feature enhancement is proposed to improve the Transformer fusion network. First, using the third- and fourth-stage features of the backbone network as inputs, channel dimension feature enhancement is performed via CAE-Net in the self-attention module of the Transformer encoder and decoder to enhance the importance of the channel. Subsequently, two-stage feature-weighted fusion and linear transformation are performed via SAE-Net to obtain the self-attention factors Q, K, and V. Finally, spatial dimension feature enhancement is performed via a self-attention operation. Experiments conducted on five widely used public benchmark datasets reveal that the improved Transformer feature fusion module can improve the tracking performance of the tracker with minimal reduction in speed of tracking.

Image segmentation is a crucial technology for environmental perception, and it is widely used in various scenarios such as autonomous driving and virtual reality. With the rapid development of technology, computer vision-based blind guiding systems are attracting increasing attention as they outperform traditional solutions in terms of accuracy and stability. The semantic segmentation of road images is an essential feature of a visual guiding system. By analyzing the output of algorithms, the guiding system can understand the current environment and aid blind people in safe navigation, which helps them avoid obstacles, move efficiently, and get the optimal moving path. Visual blind guiding systems are often used in complex environments, which require high running efficiency and segmentation accuracy. However, commonly used high-precision semantic segmentation algorithms are unsuitable for use in blind guiding systems owing to their low running speed and a large number of model parameters. To solve this problem, this paper proposes a lightweight road image segmentation algorithm based on multiscale features. Unlike existing methods, the proposed model contains two feature extraction branches, namely, the Detail Branch and Semantic Branch. The Detail Branch extracts low-level detail information from the image, while the Semantic Branch extracts high-level semantic information. Multiscale features from the two branches are processed and used by the designed feature mapping module, which can further improve the feature modeling performance. Subsequently, a simple and efficient feature fusion module is designed for the fusion of features with different scales to enhance the ability of the model in terms of encoding contextual information by fusing multiscale features. A large amount of road segmentation data suitable for blind guiding scenarios are collected and labeled, and a corresponding dataset is generated. The model is trained and tested on the dataset. The experimental results show that the mean Intersection over Union (mIoU) of the proposed method is 96.5%, which is better than that of existing image segmentation models. The proposed model can achieve a running speed of 201 frames per second on NVIDIA GTX 3090Ti, which is higher than that of existing lightweight image segmentation models. The model can be deployed on NVIDIA AGX Xavier to obtain a running speed of 53 frames per second, which can meet the requirements for practical applications.

Accurate traffic flow prediction is a key prerequisite for realizing intelligent transportation systems, and is of great significance for strengthening system simulation and control and improving the decision-making of managers. To address the problem of most existing Graph Convolutional Network (GCN) models ignoring the dynamic spatial and temporal variations in traffic data and insufficiently employing node information, which leads to insufficient extraction of spatial and temporal correlations, a traffic flow prediction model based on multiple spatio-temporal graph fusion and dynamic attention is proposed. First, the temporal characteristics of traffic flow data in multi-temporal states are extracted by different convolutional cells. The next step involves constructing a multiple spatio-temporal graph to capture the dynamic trend and heterogeneity of nodes in spatial distribution, followed by extracting spatial characteristics through the integration of GCN. Finally, the spatial and temporal characteristics are analyzed and fused using the multi-head self-attention mechanism to output prediction results. Experimental analyses are performed on two public datasets, PeMS04 and PeMS08, and compared with the Attention Based Spatial-Temporal Graph Convolutional Network (ASTGCN), Multiview Spatial-Temporal Transformer Network (MVSTT), Dynamic Spatial-Temporal Aware Graph Neural Network (DSTAGNN) and other benchmark models that utilize spatio-temporal graph convolution. The results show that the Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) of the proposed model are reduced by 7.10%, 7.22%, and 6.47%, respectively, demonstrating the proposed model′s strong adaptability and robustness.

The continuous integration of artificial intelligence and robotics technology has facilitated the widespread adoption of multi-rotor Unmanned Aerial Vehicles (UAVs) across various fields, demonstrating their flexibility and efficiency. However, when developing and validating flight control algorithms or solutions for multi-rotor UAVs, researchers face high costs and significant risks. To mitigate these risks and enhance the efficiency of algorithm testing and optimization, simulation platforms for multi-rotor UAVs provide a safe and controlled environment. In this regard, this paper first introduces conventional models of multi-rotor UAVs, selecting the commonly used quadrotor as the representative model. It then elaborates on dynamic models according to different levels of simulation. Subsequently, it provides an overview of the conventional system framework of multi-rotor UAV simulation platforms and discusses their evaluation methods and classification approaches. The evaluation of simulation platforms is detailed from the perspectives of function and performance. multi-rotor UAVs are classified based on whether they support an interactive learning environment and their focus areas: dynamics, sensors, and multi-UAV coordination. This paper also reviews the main solutions for existing UAV flight missions, analyzing typical multi-rotor UAV simulation platforms within the context of traditional and learning-based methods. Finally, the paper outlines future directions for multi-rotor UAV simulation platforms.

Object detection in Unmanned Aerial Vehicle (UAV) aerial photography images is prone to incorrect or missed detections when the target is small, obstructed, or characterized by dense scales. To address the above challenges, this paper proposes the SNA-YOLOv5s algorithm for small target detection, which is based on YOLOv5s. First, the strided convolution layer in the original model is replaced with the Spatial Depth Transformation Convolution (SPD-Conv) module, eliminating the problem of detail loss caused by strided convolution operations and enhancing the model's ability to extract features from small objects. Second, a novel Average Pyramid Pooling-Fast (AGSPPF) module is designed, and an average pooling operation layer is introduced to address the issue of information loss that occurs while extracting feature information, thereby improving the model's feature extraction capability. Third, a new large-scale detection branch specifically for small targets is added to capture rich details in shallow features and enhance the detection capability for small targets. Finally, the Normalized Attention Mechanism (NAM) is embedded in the backbone network, where feature information is weighted to suppress invalid feature information. The proposed algorithm is trained and tested on the VisDrone2019 and NWPU VHR-10 datasets, on which it achieves mean Average Precision (mAP) of 42.3% and 96.5%, respectively, which is 8.4 and 2.6 percentage points higher than that of the baseline YOLOv5s model. The robustness and accuracy of the proposed model are validated by comparisons with other mainstream deep learning models.

Deep learning has been widely applied to medical imaging. A medical image segmentation model based on an attention mechanism is one of the main methods used in current research. For the multi-organ segmentation task, most existing 2D segmentation models mainly focus on the overall segmentation effect of slices, while ignoring the loss or under-segmentation of small object feature information in slices, which limits the model′s segmentation performance. To solve this problem, this study proposes a multi-organ semantic segmentation model, DASC-Net, based on multi-scale feature fusion and an improved attention mechanism. The overall framework of the DASC-Net is based on an encoder-decoder architecture. The encoder uses the ResNet 50 network and sets a skip connection with the decoder. The attention mechanism is realized using the parallel structure of a Dual Attention Module (DAM) and a Small Object Capture (SOC) module to perform multi-scale regional feature fusion. DASC-Net not only perceives the feature information of larger objects but also retains the feature information of small objects through attention weight reconstruction, which effectively addresses the limitations of the attention module and further improves the segmentation performance of the model. The experimental results on the CHAOS dataset show that DASC-Net can obtain 83.72%, 75.79%, 87.75%, 85.63% and 77.60% on the Sensitivity, Jaccard similarity coefficient, Positivity Predictive Value (PPV), Dice similarity coefficient, and mean Intersection over Union (mIoU) indicators, respectively; the Dice similarity coefficient and 95% Hausdorff Distance (HD95) values on the Synapse dataset are 82.44% and 21.25 mm, respectively. DASC-Net performs better than the other segmentation networks on both datasets, which demonstrates its reliable and accurate segmentation performance.

Driver fatigue is a major cause of traffic accidents, and driver fatigue state classification based on Electroencephalograms (EEGs) is an important task in the field of artificial intelligence. In recent years, deep learning models that incorporate attention mechanisms have been widely applied to EEG-based fatigue recognition. While these approaches have shown promise, several studies disregard the inherent features of EEG data itself. Additionally, the exploration of the mechanisms and effects of attention on the classifier is vague, which results in failure to explain the specific effects of different attention states on classification performance. Therefore, this study selects the SEED-VIG data as the research object and adopts the ReliefF feature selection algorithm to construct optimized models of Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) network, and Support Vector Machine (SVM) based on self attention, multihead attention, channel attention, and spatial attention mechanisms. Experimental results on the EEG data included in the SEED-VIG dataset show that the performance of several neural network optimization models based on multimodal attention mechanisms has improved in terms of accuracy, recall rate, F1 score, and other indicators. Among them, the Convolutional Block Attention Module (CBAM)-CNN model, which can enhance spatial and channel information, achieves the best performance with 84.7% mean accuracy with 0.66 standard deviation.

Data privacy protection has become the focus of social attention, and countries and regions are gradually formulating relevant laws and regulations in this regard. However, because of the long and professional privacy policies released by App products, the use of automated methods to detect compliance with privacy policies has become an urgent technical challenge. Machine learning models, the widely popular solutions for this challenge, require labeled annotated datasets for support; however, a lack of such App privacy policy datasets currently exists in China. Based on the EU General Data Protection Regulation (GDPR) compliance analysis, a labeling scheme suitable for China′s Personal Information Protection Law is designed, which includes 15 required labels. Subsequently, Chinese privacy policies for 363 Apps in 10 categories are obtained using Web crawlers, and these privacy policies are classified and annotated at the sentence level. A Chinese privacy policy corpus consisting of 104 134 privacy policy statements and labels is constructed. The corpus is trained and tested using the latest open-source pretraining language model from Baidu, ERNIE, with a detection accuracy of 85.75%.

Owing to the heterogeneity and complexity of Android malware, traditional static analysis methods that rely on single features such as permissions or API often struggle to accurately differentiate between benign and malicious applications. To address this limitation, this study proposes a novel feature construction method based on multi-modal feature fusion based on in-depth research of Android software features such as permissions, API, bytecodes, and opcodes. The bytecode is transformed into RGB images and visual representations are extracted using the pretrained EfficientNetV2B3 model to capture the high-level characteristics of Android applications. Additionally, Locality-Sensitive Hashing (LSH) is employed to extract opcode sequence features that represent low-level, detailed characteristics of the application. These heterogeneous features are then fused using a Multimodal Factorized Bilinear pooling (MFB) algorithm to create a more discriminative representation of the malware. Building on this enhanced feature representation, a Transformer Encoder-based Android Anomaly Detection (TEAAD) model is introduced. By leveraging the transformer architecture, the TEAAD effectively learns to detect anomalies in Android malware. The experimental results demonstrate that the TEAAD model based on fused features outperforms other deep-learning models, achieving a detection accuracy of 96.87%. The MFB feature fusion method exhibits superior malware identification capabilities compared with other research methods.

When implementing Federated Learning (FL) in a cell-free network environment, user scheduling and resource allocation strategies are crucial for optimizing system time overhead, improving user reachability, and accelerating FL convergence rate. To address the issue of uneven resource allocation, this study designs an optimization scheme that combines user scheduling, CPU processing frequency, and power allocation. This scheme aims to achieve fair resource allocation by maximizing the minimum user rate in the system, thus enhancing FL performance. The joint optimization problem is decomposed into two subproblems: user scheduling and power allocation. For user scheduling, this study proposes a greedy scheduling algorithm based on k-means clustering to comprehensively evaluate channel conditions and data "value" of users and categorize users into different groups. Subsequently, for the resource occupation situation, a personalized CPU processing frequency allocation plan is developed for users within each group based on their resource occupancy. Finally, by independently executing user scheduling within each group, user selection is performed efficiently and precisely, and the complexity of user selection is effectively reduced via early grouping. For power allocation, this study introduces a Bisection Method-based Power Allocation (BM-PA) algorithm. This algorithm not only considers fairness among users but also prioritizes resource-constrained users to ensure that they can obtain superior resource allocation. The BM-PA algorithm achieves fast convergence of power allocation using a low-complexity iterative optimization process, significantly improving the resource utilization efficiency without deteriorating the system performance. In this study, a reasonable user scheduling strategy serves as the foundation for obtaining optimal solutions for the power allocation subproblem. This study adopts an alternating iteration method that allows independent optimization in each subproblem while considering the solution of the other subproblem. Via multiple rounds of iterative optimization, this interdependent relationship ensures that power resources are reasonably allocated to users who need them the most or are most likely to effectively utilize them, thus enhancing the overall system performance. This study realizes joint optimization solutions that significantly improve overall system performance. Simulation results show that compared with the baseline algorithm, the proposed algorithm exhibits outstanding performance in terms of downlink achievable rates-the average improvement reaches up to 103.34% under optimal conditions. Additionally, the uplink achievable rates improve by up to 102.78%. Furthermore, the proposed algorithm can save 67.44% of the FL task training time on average compared to the baseline algorithm, particularly when the FL learning model accuracy reaches 90%, wherein the time overhead of the proposed algorithm is minimal.

The generation of High-Definition (HD) environmental semantic maps is indispensable for environmental perception and decision making in autonomous driving systems. To address the modality discrepancy between cameras and LiDARs in perception tasks, this paper proposes an innovative multimodal fusion framework, HDMapFusion, which significantly improves semantic map generation accuracy via feature-level fusion. Unlike traditional methods that directly fuse raw sensor data, our approach innovatively transforms both camera images and LiDAR point cloud features into a unified Bird's-Eye-View (BEV) representation, enabling physically interpretable fusion of multimodal information within a consistent geometric coordinate system. Specifically, this method first extracts visual features from camera images and 3D structural features from LiDAR point clouds using deep learning networks. Subsequently, a differentiable perspective transformation module converts the front-view image features into a BEV space and the LiDAR point clouds are projected into the same BEV space through voxelization. Building on this, an attention-based feature fusion module is designed to adaptively integrate the two modalities using weighted aggregation. Finally, a semantic decoder generates high-precision semantic maps containing lane lines, pedestrian crossings, road boundary lines, and other key elements. Systematic experiments conducted on the nuScenes benchmark dataset demonstrate that HDMapFusion significantly outperforms existing baseline methods in terms of HD map generation accuracy. These results validate the effectiveness and superiority of the proposed method, offering a novel solution to multimodal fusion in autonomous driving perception.

Detecting defects on unstructured roads is important for road traffic safety; however, annotated datasets required for detection is limited. This study proposes the Multi-Augmentation with Memory (MAM) semi-supervised object detection algorithm to address the lack of annotated datasets for unstructured roads and the inability of existing models to learn from unlabeled data. First, a cache mechanism is introduced to store the positions of the bounding box regression information for unannotated images and images with pseudo annotations, avoiding computational resource wastage caused by subsequent matching. Second, the study proposes a hybrid data augmentation strategy that mixes the cached pseudo-labeled images with unlabeled images inputted into the student model, to enhance the model′s generalizability to new data and balance the scale distribution of images. The MAM semi-supervised object detection algorithm is not limited by the object detection model and better maintains the consistency of object bounding boxes, thus avoiding the need to compute consistency loss. Experimental results show that the MAM algorithm is superior to other fully supervised and semi-supervised learning algorithms. On a self-built unstructured road defect dataset, called Defect, the MAM algorithm achieves improvements of 6.8, 11.1, and 6.0 percentage points in terms of mean Average Precision (mAP) compared to those of the Soft Teacher algorithm in scenarios with annotation ratios of 10%, 20%, and 30%, respectively. On a self-built unstructured road pothole dataset, called Pothole, the MAM algorithm achieves mAP improvements of 5.8 and 4.3 percentage points compared to those of the Soft Teacher algorithm in scenarios with annotation ratios of 15% and 30%, respectively.

In response to the increasing demand for fast response and large-scale coverage in application scenarios such as satellite data processing and vehicle remote control, this study focuses on utilizing hierarchical control and artificial intelligence technology to design a resource scheduling mechanism for space-air-ground integrated computing power networks. Air, space, and ground networks are divided into three domains, and domain controllers are deployed for resource management in the corresponding local domain. The areas are divided based on the coverage of satellites and drones to ensure that they can achieve effective service guarantees, efficient data transmission, and task processing. A multi-agent reinforcement learning-based scheduling algorithm is proposed to optimize resource utilization in space-air-ground integrated computing power networks, considering each domain controller is treated as an agent with task scheduling and resource allocation capabilities. Intelligent resource scheduling and efficient resource allocation for computing tasks are realized through collaborative learning and distributed decision-making with satisfactory delay and energy consumption constraints. Computing tasks are generated in different scenarios and processed in real time. Simulation results show that the proposed mechanism can effectively improve resource utilization and shorten task response time.

To address the challenges of insufficient computing capacity of end users and the unbalanced distribution of computing power among edge nodes in computing power networks, this study proposes an Unmanned Aerial Vehicle (UAV)-assisted Device-to-Device (D2D) edge computing solution based on incentive mechanisms. First, under constraints involving computing resources, transmission power, and the unit pricing of computing resources, a unified optimization problem is formulated to maximize system revenue. This problem aims to optimize the task offloading ratio, computing resource constraints, UAV trajectory, as well as the transmission power and unit pricing of computing resources for both UAVs and users. The Proximal Policy Optimization (PPO) algorithm is employed to establish user offloading and purchasing strategies. In addition, an iterative strategy is implemented at each time step to solve the optimization problem and obtain the optimal solution. The simulation results demonstrate that the PPO-based system revenue maximization algorithm exhibits superior convergence and improves overall system revenue compared to the baseline algorithm.

For improving system security and stability, multivariate time series anomaly detection methods are generally used to detect and identify abnormal patterns and behaviors in systems. To mitigate reduced anomaly detection accuracy caused by complex dependencies within multivariate time series, a multivariate time series anomaly detection model, HGAT, is proposed. This model is based on a graph attention network and is optimized by combining prediction and reconstruction methods. First, graph attention networks are used to capture the temporal and spatial dependencies of multivariate time series. Second, a Transformer that integrates Variational Autoencoders (VAE) is used as the reconstruction module and a Time Convolutional Network (TCN) is used as the prediction module to jointly detect abnormal sequences. The self-attention mechanism of the Transformer enables the reconstruction module to model the entire time series by directly considering the relationship between any two positions in the sequence to capture the global dependency relationship of the sequence. TCN can effectively extract local features of the time series by stacking convolutional layers and increasing the receptive field. Finally, by comprehensively considering the reconstruction and prediction modules via abnormal scores, the overall distribution of the sequence is analyzed from both global and local perspectives based on spatiotemporal joint representation. Experiments are conducted on the SMAP, MSL, and SMD datasets, and the results show that the F1 values for the HGAT are 96.20%, 97.50%, and 92.85%, respectively. These values are superior to those for the baseline method.

With the rapid advancement of general artificial intelligence technology, the application of foundational models across various fields has gained increasing attention. In image segmentation, the Segment Anything Model (SAM), as a foundational model, demonstrates notable advantages in enhancing image comprehension and processing efficiency. While SAM achieves state-of-the-art performance in image segmentation, further optimization in power consumption, computational efficiency, and cross-domain adaptability is required. This review provides an in-depth exploration of the potential improvements to SAM across several crucial dimensions, such as enhancing speed and computational efficiency, improving model accuracy and robustness, increasing adaptability and generalization, optimizing prompt engineering, and boosting data utilization and transfer learning capabilities. With these enhancements, SAM is expected to sustain high efficiency in highly complex tasks and better meet requirements of various fields and application contexts. In addition, this review summarizes the practical applications of SAM in various fields, including medical imaging, remote sensing, and the mechanical industry, and demonstrates the suitability and challenges of the model in different scenarios. Moreover, this review provides a detailed overview of commonly used datasets and evaluation metrics in the field of image segmentation. Through experimental comparative analyses, the impact of Vision Transformer (ViT) variants on the performance of SAM is assessed, along with performance evaluations of enhanced models, such as EfficientSAM, EfficientViT-SAM, MobileSAM, and RobustSAM. The challenges faced by SAM and its improved models in real-world applications are also discussed, and future research directions are proposed. This review aims to provide researchers with a comprehensive understanding of the advancements and applications of SAM and its variants, offering insights that may inform the development of new models.

The complex backgrounds, diverse target types, and significant scale variations in remote sensing images lead to target omission and false detection. To address these issues, this study proposes an improved Faster R-CNN multi-object detection model. First, the ResNet 50 backbone network is replaced with the Swin Transformer to enhance the model's feature extraction capability. Second, a Balanced Feature Pyramid (BFP) module is introduced to fuse shallow and deep semantic information, further strengthening the feature fusion effect. Finally, in the classification and regression branches, a dynamic weighting mechanism is incorporated to encourage the network to focus more on high-quality candidate boxes during training, thereby improving the precision of target localization and classification. The experimental results on the RSOD dataset show that the proposed model significantly reduces the number of Floating-Point Operations per second (FLOPs) compared to the Faster R-CNN model. The proposed model achieves 10.7 percentage points improvement in mAP@0.5 ∶0.95 and 10.6 percentage points increase in Average Recall (AR). Compared to other mainstream detection models, the proposed model achieves higher accuracy while reducing the false detection rate. These results indicate that the proposed model significantly enhances detection accuracy in remote sensing images with complex backgrounds.

In traffic sign detection, external environmental interference and the small size of traffic sign targets in driving scenarios hinder detection performance. This paper introduces a novel traffic sign detection algorithm that significantly improves model detection precision while ensuring real-time detection capabilities. This paper initially designs a new multi-scale feature extraction network, incorporating large-scale features to augment small target localization information, and simultaneously designs a multi-scale feature attention enhancement module to further enhance the model′s feature extraction capability. Second, to reduce the computational load and complexity of the model, this paper improves the multi-scale detection heads of the original model by selecting two large-scale detection heads for detecting small targets. Finally, the algorithm modifies the Complete Intersection over Union (CIoU) loss function to enhance its perception of small targets and improve the network′s training efficiency. On two open-source public datasets, namely the TT100K and CCTSDB 2021 traffic sign datasets, the improved model achieves enhanced detection precision for small traffic sign targets, with a mean Average Precision (mAP) of 84.8% and 83.6% on the test sets, respectively. These results show improvements of 3.0 and 3.6 percentage points over the baseline models, respectively, demonstrating the model′s higher detection performance and feature extraction capabilities while meeting the requirements for real-time detection.