|
WELCOME TO |
| Acta Electronica Sinica |
|
Acta Electronica Sinica, launched in 1962, is the flagship academic journal of the Chinese Institute of Electronics (CIE). The journal publishes cutting-edge research findings in the fields of electronics, information science, and related disciplines.
|
| LEARN MORE |
Quick Entry
- Ex-Submission System
- Submission System
- Contact Us
 |
 |
 |
Articles Online
- 摘要：In recent years, with the rapid advances in multimodal pre-trained models, image description generation has progressed from brief and general descriptions to more specific, detailed, and flexible content expressions. However, existing image description models still mainly focus on surface-level visual content, such as object categories, appearance attributes, and spatial locations, making it difficult to adequately satisfy users' needs for a more comprehensive understanding in real-world applications, including fine-grained category information, functional purposes, and relevant external background knowledge. Particularly in knowledge-intensive vertical domains such as museum guiding and artwork interpretation, users expect not only shallow visual descriptions of images, but also additional in-depth information, such as historical origins, artistic schools, symbolic meanings, and other related content. In addition, explicitly aligning narrative content with key visual regions in an image and simultaneously grounding textual segments to their corresponding image regions can help users understand the relevant knowledge more intuitively and support further interaction and exploration around specific details. Therefore, this paper introduces the task of Grounded Image Narrative Generation, which takes an image and a user-provided multimodal instruction as input and generates a multimodal narrative containing rich information, coherent logical relationships, and accurate region grounding. To achieve this goal, we propose a knowledge retrieval-augmented narrative generation method that first actively retrieves rich background information from external knowledge bases. We then introduce a chain-of-thought reasoning mechanism jointly constrained by cross-modal knowledge association and region localization, which selects effective knowledge relevant to the user query, and leverages visual localization to better align the generated text with corresponding local image regions, ultimately enabling decomposition and reasoning of queries as well as structured generation of narrative content. For data construction, we extend the encyclopedic image narrative dataset ImgNarr-23K with region-grounding annotations and further construct ArtNarr, a dataset specifically for the artwork domain, providing data support for research on knowledge-intensive image narrative generation. Experimental results show that the proposed method outperforms comparison methods based on mainstream multimodal large models across datasets from multiple domains. It is capable of generating high-quality image narratives with integrated textual and visual content, containing complete and accurate image information and relevant knowledge while achieving effective region grounding, thereby demonstrating its strong potential and practical value in real-world multimodal application scenarios.
36
6
0
2026-09-22
- 摘要：To address the increased scan-gain fluctuation in millimeter-wave dual-polarized phased arrays caused by the mismatch between the element radiation pattern and the array-factor envelope during wide-angle scanning, a design method for wide-angle phased arrays with low scan-gain fluctuation based on an externally loaded dielectric pin array (DPA) is proposed. First, based on the concept of pattern complementation, a planar DPA is employed to weakly load and spatially redistribute the near field above the dual-polarized antenna element. As a result, a wide-beam element pattern with a center dip is formed, which enhances the radiation at large off-boresight angles and suppresses the gain degradation during wide-angle scanning. Then, a 1 × 8 millimeter-wave dual-polarized phased-array prototype loaded with the planar DPA is designed, fabricated, and measured. The measured results show that the DPA can extend the scanning range and reduce the scan-gain fluctuation. Furthermore, to alleviate the broadside gain reduction introduced by the planar DPA, an arched DPA structure with a lens-like arrangement is proposed. By introducing continuous spatial dielectric loading, the aperture phase distribution is improved, thereby enhancing the broadside radiation performance while maintaining the gain-compensation capability at large scanning angles. This enables a better gain balance between the broadside and large-angle scanning states. Simulated results show that, after loading the arched DPA, the array can achieve wide-angle scanning of approximately ± 60° for both polarizations over the frequency range of 24~29 GHz, and no obvious gain degradation is observed at the scanning endpoints compared with the broadside gain. This indicates that the proposed structure can effectively improve the gain stability during wide-angle scanning. The proposed DPA is an externally loaded structure with the advantages of low profile, simple configuration, easy fabrication, and convenient integration with PCB-based millimeter-wave phased arrays. The proposed method can provide a reference for the low scan-gain fluctuation design of millimeter-wave wide-angle phased arrays and has potential engineering applications in millimeter-wave communication, radar detection, and other array systems requiring wide-angle coverage.
14
11
0
2026-09-22
- 摘要：In order to solve the problems of limited resources and low utilization rate caused by high-speed satellite mobility and dynamic changes of ground services in Low Earth Orbit (LEO) satellite networks, this paper studies the joint optimization method for uplink Random Access (RA) and Data Transmission (DT) in mixed service scenarios. By analyzing the different requirements of Real Time (RT) service and Best Effort (BE) service for latency and bandwidth, the unit step function is used to represent the resource requirements of RT service, and the exponential utility function is used to represent the elastic demand of BE service for resources. Based on the utility function of two services, a network utility maximization model for the coexistence of mixed services is established. Due to the fluctuation of ground traffic load in different periods, in order to solve this problem, this paper introduces a traffic load prediction mechanism based on Method of Moments (MOM). This mechanism mainly estimates active users by observing the number of idle, successful access and conflicting access opportunities. Based on this prediction mechanism, an iterative resource scheduling algorithm is designed, which is based on the predicted load to dynamically adjust the access control dianprobability and the time-frequency Resource Blocks (RBs) allocation scheme in the RA and DT stages. The simulation results show that the performance of the proposed method under different traffic loads and network conditions is close to the theoretical optimal value. In comparison with the fixed resource proportion benchmark method, the proposed method can increase the average utility of the network by about 37%, 17% and 9%, respectively, compared with the benchmark method under different proportion settings. At the same time, in the multi-cycle dynamic traffic scenario, the average utility is further increased by 57%, 26% and 10%. The simulation results show that this method can not only effectively reduce the collision probability in the RA phase, ensure the quality of RT service and improve the throughput of BE service, but also significantly improve the overall Quality of Service (QoS) and resource utilization of LEO satellite network.
10
7
0
2026-09-22
- 摘要：RNA sequence design and generation serve as core technologies for the development of functional RNA elements. To address the key challenges in current RNA sequence generation, including insufficient sequence representation capability, difficulties in structural constraint modeling, and low quality of generated sequences, this paper proposes a variational autoencoder (VAE) integrating context-free grammar and semantic embedding, termed CFG-Embed-RNAgenor. The proposed method constructs an extended G5 grammar that supports gap character processing. Based on the extended G5 grammar, RNA sequences and their corresponding secondary structures are converted into parse trees, from which the production rule sequences are derived. Subsequently, integer encoding is performed on the rule sequences to realize the joint modeling of RNA sequence information and secondary structure constraints. On this basis, a VAE architecture integrating convolutional neural networks and semantic embedding is designed. Replacing the traditional one-hot encoding, the introduced semantic embedding captures the intrinsic semantic correlations among production rules, thereby effectively improving the model's capability of representing RNA sequence features. Comparative experiments conducted on ten RNA families from the Rfam database demonstrate that the proposed method achieves a substantial improvement in the bit score of generated sequences compared with the state-of-art method RNAgg, verifying its superiority in high-quality RNA sequence generation. Ablation experiments validate the effectiveness of each core module: the feature learning strategy combining CNN and semantic embedding can better enhance the quality of generated sequences; both sequence alignment information and RNA structural information positively contribute to model generation performance, among which sequence alignment information exhibits a more prominent optimization effect. Meanwhile, the latent space of the model can effectively encode the phylogenetic information of RNA families. Furthermore, the introduction of a temperature sampling mechanism enables flexible regulation of the diversity and consistency of generated sequences, effectively balancing the overall generation performance of the model. This study provides a novel computational method for the precise design of RNA family sequences and offers new insights into the joint representation learning of RNA sequences and structures.
12
7
0
2026-09-22
- 摘要：The test case prioritization (TCP) technologies in continuous integration (CI) environments are critical approaches to accelerating defect detection and shortening feedback cycles. However, existing machine learning-based TCP methods are limited to the individual-level dimension of test cases in their feature engineering, failing to season to the organizational hierarchy inherent in the multi-stage pipeline structures of industrial-level CI environments, resulting in a lack of structural context. Meanwhile, these models lack a mechanism for evaluating feature importance relative to the current cycle during training and prediction, preventing them from dynamically selecting key features that align with the prediction signals of the current cycle. Consequently, the models lack the ability to effectively adapt to failure patterns specific to the current cycle. To address both issues, we propose MCCD-TCP, a test case prioritization approach in continuous integration based on multi-dimensional context feature and cycle-aware dual model. Firstly, we construct multi-dimensional contextual features that encompass execution time distribution, pipeline stage risk, and test class association, expanding the feature space from individual attributes to the organizational level to provide the model with failure prediction dimensions beyond individual attributes. Secondly, we design a cycle-aware dual model collaborative ranking mechanism that adaptively selects key features for the current cycle. We train two separate random forest models using feature vector sets composed of global features and key features, respectively, and perform weighted fusion for prediction to obtain the optimal prioritized test sequence. This approach enhances the model’s adaptability to dynamic CI testing environments while balancing ranking stability and alignment with cycle-based failure patterns. Finally, to validate the practical value of the proposed method in complex CI scenarios, a comprehensive evaluation is conducted on the LRTS dataset, which covers 10 large-scale open-source projects and features a real-world multi-stage pipeline structure. Comparative experimental results demonstrate that the proposed approach is feasible and effective, achieving average improvements of 1.21% and 1.22% on the APFD and APFDc metrics, respectively, compared to existing classical and recent state-of-the-art methods. Specifically, on the Hadoop and James projects, our approach outperforms the best-performing baseline by 2.28% and 2.19% for APFD, and by 2.05% and 3.36% for APFDc, respectively. Meanwhile, the RFTC metric results further demonstrate the method’s ability to detect early-stage failures. Ablation experiments further show that the combined design of multi-dimensional context feature extraction and cycle-aware random forest dual model ranking is effective, offering significant advantages over individual variant models. Furthermore, cold-start analysis indicates that the performance gain from multi-dimensional context features gradually emerges and stabilizes as historical data accumulates, and the overall computational overhead of the proposed approach remains manageable.
11
9
0
2026-09-22
- 摘要：Hash functions are fundamental primitives in modern cryptography, and their security is typically characterized by three essential properties: preimage resistance, second preimage resistance, and collision resistance. Hash combiners, which compose multiple hash functions in parallel or in series, aim to maintain a high level of security. Consequently, they are widely deployed in protocols such as TLS and SSL. Compared with the traditional random oracle model, the weak ideal compression model imposes weaker security assumptions, thereby making the security analysis of such combiners more demanding and more practically relevant. In recent years, with the development of quantum computing, in particular the square-root speedup offered by Grover's algorithm for unstructured search, the security boundaries of these constructions have been significantly challenged, and it is imperative for the community to reassess the security of existing hash combiners under quantum adversary models. Within the security analysis of hash combiners (XOR combiner, Zipper, and Hash-Twice) based on weak ideal compression functions, this paper proposes a preimage attack on Hash-Twice in the classical setting, by constructing a multicollision-and-double-diamond structure using Joux's multicollision technique and optimizing the involved parameters. Considering two scenarios: one where a substantial amount of cheap quantum random access memory (qRAM) is available, and another where qRAM is limited and expensive, this paper extends the inverse-diamond structure and multicollision-and-double-diamond structure from the classical setting to the quantum setting, using quantum multicollision techniques based on the BHT or CNS algorithms. For the preimage attack on Hash-Twice and the existing second preimage attacks on XOR combiner and Zipper, quantum inverse-diamond structure and quantum multicollision-and-double-diamond structure are employed, combined with quantum collision search, to transform them into general attacks in the quantum setting. We also analyze the cost of these attacks. Based on the aforementioned techniques, this work demonstrates that in the classical setting, when the underlying compression function is modeled as a weak ideal compression function, the time complexity of preimage attacks on double hashing is
19
6
0
2026-09-22
Featured Articles
PAPERS
摘要：With the explosive growth of image data in the field of e-commerce, target image retrieval has become a challenging work in information retrieval research. The existing traditional image retrieval models only rely on a single text description or similar image, which is difficult to accurately capture the user’s retrieval intention, resulting in unsatisfactory retrieval results. In order to solve this problem, this paper proposes a combined retrieval method that fuses image and text features. Swin Transformer (SwinT) is used to extract the multi-layer features of the reference image, and the image and text features are fused at multiple levels, so that the text features can modify the reference image features at multi-level and fine-grained, and get closer to the target image features. Then, the modified image features and the target image features are embedded in a space for similarity measurement, and the batch-based classification loss is used to optimize the retrieval performance. Experimental results on Fashion200k, MIT-States and CSS datasets show that the proposed method improves the performance by 5 percentage points on average compared with the existing mainstream methods.1181
397
2
2025-12-08
PAPERS
摘要：In unsupervised domain adaptation tasks, the source and target domains usually do not satisfy the independent and identical distribution assumption. In order to generate the usable labels for the target domain, classical domain adaptation methods select the category with the highest prediction probability of the classifier as the pseudo-label of the target sample. Thus, the pseudo-label inevitably contains certain noise information, which may cause negative transfer to the domain adaptation model. In addition, traditional adversarial domain adaptation methods usually consider the global distribution between domains and ignore the category information of samples. How to extract discriminative category-level features in domain adaptation tasks is also an important problem. Therefore, an unsupervised adversarial domain adaptation method is proposed using feature anomaly detection and pseudo-label regression. The target samples of the same class predicted by the classifier are formed into the category subdomain within the target domain. The Gaussian uniform mixture model is used to detect the subdomain samples with abnormal distance from the class mean. The posterior probability of the samples is calculated and the correctness of the sample pseudo-labels in the subdomain is measured, which is used as a loss factor to limit the influence of pseudo-labels on the model in training. Meanwhile, the pseudo-label regression function is used to reduce the difference between the predicted label and the high-confidence pseudo-label of the classifier. The category constraint of the unlabeled target domain is adopted to improve the distinguishability of feature categories. Experimental results show that the average recognition accuracy of the proposed method on datasets Office-31, Image-CLEF, and Office-Home are 90.2%, 89.6%, and 69.5%, respectively, which are all higher than the related popular algorithms.651
337
3
2025-12-08
Special Focus: Recipient of CIE Outstanding Doctoral Dissertation
摘要：This paper proposes a contactless human vital-sign sensing method in non-line-of-sight (NLoS) scenarios, which integrates reconfigurable intelligent surfaces (RISs) into wireless human sensing, and achieves NLoS human vital-sign sensing by utilizing the flexible manipulation capability of RISs over electromagnetic waves. Firstly, through a visual-aided mechanism, this paper locates human targets using the deep learning model Yolo-v7. Then, the metasurface optimizes the coding matrix based on the estimated human location, thereby altering the propagation path of sensing signals to achieve human sensing in non-line-of-sight scenarios. Additionally, addressing the shortcomings of traditional variational mode decomposition (VMD) algorithms in human vital-sign sensing, we further propose an improved VMD algorithm to achieve precise estimation of breathing rate and heartbeat rate. Experimental results demonstrate that the visual-aided module can accurately locate the position of human chest, thus enabling precise beamforming controlled by the intelligent metasurface. The proposed filtering-based VMD algorithm can achieve precise estimation of human vital signs, with average estimation errors of 0.6 RPM for respiration and 5.3 BPM for heartbeat, respectively. Furthermore, further analysis also demonstrates the effectiveness and accuracy of the RIS-based sensing scheme for human vital-sign sensing in NLoS scenarios.677
363
1
2025-12-08
PAPERS
摘要：The wireless powered communication is a key technology to support energy sustainable wireless communications in the future 6G era. For the first time, this paper proposes a complete prototype design for the large-scale antenna based wireless powered communication system, and realizes this prototype system, which consists of a base station with a 32-transmit-antenna array, a user equipment and a feedback link, on the software defined radio platforms. At the base station, by designing flexible frame structures, this paper originally proposes a dynamic wireless energy transmission scheme to properly balance the required channel estimation accuracy and the wireless energy transmission efficiency. At the user equipment, by applying the practical non-linear energy harvesting and energy consumption models, this paper proposes a novel information-energy double threshold based battery management scheme to meet the requirements of different information and energy transmission. Based on the prompt status feedback from the user equipment, this paper completes a close-loop controlled system framework for adaptive wireless information and energy transmissions. Extensive experiments for both indoor and outdoor scenarios are implemented to verify the good environment adaptation and transmission performance of the wireless powered communication prototype system.434
311
0
2025-12-11
PAPERS
摘要：With the rapid development of the fifth generation (5G) mobile communications technology, Internet of vehicles (IoV) has become one of the important scenarios for 5G applications. Due to the urgent requirements of these applications for timely content caching, a dynamic content caching-enabled edge caching scheme is investigated to ensure timeliness, which allows vehicles to obtain cache content with freshness constraints from an edge caching-enabled access point (CAP). Furthermore, based on the idea of collaborative cache, a random model of CAP cache refreshing and delivery is established by probability theory. Meanwhile, considering the metric of hit rate with age of information (AoI) as the quality of service (QoS) index of IoV cache, a system model of caching performance value is developed. To ensure the effectiveness of content caching, two types of vehicular content requests with different AoI requirements are considered in a hierarchical slicing framework. Finally, the optimization problem to maximize the content caching value of cooperative content placement is formulated, which is a multi-objective, multi-dimensional and multi-choice knapsack problem and is proved to be NP-hard. Therefore, the problem is divided into several independent optimization problems, and a maximum incremental algorithm based on the age value of the content is designed to solve it. Simulation results show that the CAP cooperative content placement strategy outperforms other traditional methods.371
294
3
2025-12-11
PAPERS
摘要：At present, there are few researches on profile attacks against RSA-CRT implementation. This paper takes modular reduction operation as the research object, and a template attack method against RSA-CRT implementation is proposed. The core of this method is to solve the difficulty to recover the RSA-CRT private key from the Hamming weight of the intermediate value of ciphertext modular reduction. The characteristic of this method is to build a model based on the Hamming weight of the intermediate value derived from modular reduction. The Hamming weight can be obtained by collecting the power traces of chosen ciphertext modular reduction for template matching, and the intermediate value is recovered from the Hamming weight variation, the private key of the RSA-CRT algorithm can be further inferred based on the intermediate value. In addition, the advantage of this method is that ideally, templates based on the intermediate Hamming weight model can be shared, and there is no limit on the number of bits of the intermediate value for modelling, which can be in byte size, 64 bit size, or even the bit size of p. In the actual environment, it can be selected according to the leaked information. Finally, in this paper, the lowest byte of the intermediate value is selected to model to verify the feasibility of this method, and the defense suggestions are also provided.324
267
0
2025-12-11
0 1 Global Dependency Guided Feature Reconstruction for Image Classification
0 2 Radar-Infrared Multi-View Cooperative Intelligent TBD Technology for Maritime Weak Target Detection
0 4 A Review of Vision-Based UAV Localization and Navigation Methods
0 6 CIMOT3D: Chinese-Instruction-Based Monocular 3D Multi-Object Tracking
0 7 Systems and Key Technologies of 4K/8K UHD TV Production Broadcasting and Presentation
0 8 Fusing Large Language Models with Cross-Domain Graphs for Multimodal Emotion Recognition in Dialogue
查看更多
Virtual Issues
查看更多
| Multi-media Resources |
| Read about our articles with data, invited articles, videos ...... |
媒体矩阵
icon
icon
icon
icon
icon
icon
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0