Table of Content

25 July 2026, Volume 2026 Issue 4 Previous Issue   

For Selected:

Download Citations EndNote Reference Manager ProCite BibTeX RefWorks

Toggle Thumbnails

Select

Contrast structure in a singularly perturbed Burgers type equation Xiao WU, Yue LIU 2026, 2026 (4):  1-12.  doi: 10.3969/j.issn.1000-5641.2026.04.001 Abstract ( 0 )   HTML ( 1 )   PDF (1216KB) ( 0 )   Save This study considers the periodic boundary value problem for a class of singularly perturbed Burgers type equation with a discontinuous reaction term. Based on the boundary layer function method, contrast structure theory, and differential inequality methods, the existence and uniqueness as well as the local stability of the contrast structure solution are proved, and a high-precision asymptotic expansion of the solution is established. Figures and Tables | References | Related Articles | Metrics

Select

Polarization dynamics of photoluminescence from CsPbBr3 submicron spherical microcavities Wei SUN, Wei XIE 2026, 2026 (4):  13-20.  doi: 10.3969/j.issn.1000-5641.2026.04.002 Abstract ( 0 )   HTML ( 0 )   PDF (1219KB) ( 0 )   Save CsPbBr3, an all-inorganic perovskite, has emerged as a popular material in the photoelectric field because of its excellent photoelectric characteristics. This study focuses on the CsPbBr3 structure, containing submicron spherical cavities, revealing the microscale interactions between photons and excitons. By analyzing the photoluminescence spectra, radiation lifetime, and polarization evolution law at different excitation powers, it was discovered that the CsPbBr3 microsphere cavity achieved high circular polarization lasing above the lasing threshold. Time-resolved spectroscopy was used to analyze the temporal polarization characteristics of CsPbBr3 submicron spheres in detail during the radiation process, and the lasing peak exhibits a degree of circular polarization as high as 0.65. These results provided a foundation for the design and development of new photoelectric devices using chiral quantum optics. Figures and Tables | References | Related Articles | Metrics

Select

Study on sensitivity improvement of unbalanced interferometers through optical amplification Tingting JIANG, Yuan WU, Liqing CHEN 2026, 2026 (4):  21-28.  doi: 10.3969/j.issn.1000-5641.2026.04.003 Abstract ( 0 )   HTML ( 0 )   PDF (1223KB) ( 0 )   Save The Mach-Zehnder interferometer is a key tool for precision measurement, and its phase sensitivity is critical to measurement accuracy. However, photon loss limits the increase in phase sensitivity. Although adjusting the interferometer’s splitting ratio can reduce the impact of unbalanced loss, increasing absolute sensitivity remains an unresolved issue. To compensate for photon loss, this study introduces an amplifier into a Mach-Zehnder unbalanced lossy interferometer with an adjustable splitting ratio. Theoretical analysis shows that this scheme can significantly improve phase sensitivity, with more pronounced enhancement effects observed as loss increases, exhibiting good adaptability and robustness. Compared with increasing input light intensity, this scheme provides a novel approach to improving interferometer performance. Figures and Tables | References | Related Articles | Metrics

Select

Video deblurring model based on frame attention Mengjie LU, Lei CHEN 2026, 2026 (4):  29-41.  doi: 10.3969/j.issn.1000-5641.2026.04.004 Abstract ( 0 )   HTML ( 0 )   PDF (8176KB) ( 0 )   Save Existing video deblurring models can be categorized into two types: sliding window-based and loop-based methods. Sliding window-based methods process consecutive frames as input, limiting the clear information available to the central frame. Loop-based methods use a recurrent neural network to process each video frame sequentially, but the implicit features carrying clear information gradually weaken as the propagation distance increases. To address these issues, this article proposes a frame attention-based method—a novel video deblurring framework that differs from the two aforementioned methods by selecting the most suitable set of auxiliary frames for the central frame. To mitigate the challenge of large displacement between frames, we designed a lightweight optical flow model and fine-tuned it using distillation to enhance the alignment module’s ability to process blurred images. Experiments on two commonly used video deblurring datasets demonstrate that our method achieves a peak signal-to-noise ratio of 31.91 dB, indicating strong performance. Figures and Tables | References | Related Articles | Metrics

Select

Gradient boosting decision tree based federated learning framework Yao LIU, Runmeng DU, Lei CHEN 2026, 2026 (4):  42-50.  doi: 10.3969/j.issn.1000-5641.2026.04.005 Abstract ( 0 )   HTML ( 0 )   PDF (2077KB) ( 0 )   Save This study proposed an efficient and secure federated learning framework based on gradient boosting decision tree. It was used to protect privacy within a vertical federated learning environment and was capable of handling situations where feature scales were completely different or where binary and continuous features coexisted. This framework employed LightGBM as the boosting decision tree and used a symmetric encryption mechanism to safeguard gradient privacy. Security analysis indicated that it did not disclose gradient privacy and was effective in defending against the risks of data eavesdropping or tampering during transmission. Experiments conducted on biomedical datasets and commonly used credit prediction datasets validated the effectiveness of this framework, demonstrating higher efficiency than other existing gradient boosting decision tree based federated learning frameworks while maintaining the same level of accuracy. Figures and Tables | References | Related Articles | Metrics

Select

Deep learning-based method for automatic verification of platen status in distribution cabinet Ning CHEN, Rongsheng LIN, Cheng YUAN, Jin SHANG, Fan BAI, Dingjiang HUANG 2026, 2026 (4):  51-62.  doi: 10.3969/j.issn.1000-5641.2026.04.006 Abstract ( 0 )   HTML ( 0 )   PDF (4470KB) ( 0 )   Save In a substation, the platen is a core component on the distribution cabinet. In case of failure or unintentional human operation, the state verification of the platen of the distribution cabinet is important in the manual inspection of the substation as platen is prone to misinvestment and misreturn. Therefore, a new deep learning-based automatic checking method is proposed for the status of switchgear platen. First, the template platen frame data are extracted from the Excel-formatted panel diagram, in which every switchgear cabinet corresponds to a panel diagram. The panel diagram is an electronic record of the components in the switchgear cabinet, which contains the correct casting and retiring status of each platen, the name of the platen, and other information, i.e., the template platen frame data. Then the YOLOv5s algorithm is used to detect the platen target on the captured image to obtain the predicted platen detection frame data. The connectionist temporal classification (CTC) probability of paddle-to-paddle optical character recognition (PP-OCRv4) is used as the text similarity measure between the predicted and stencil platen names. Finally, based on the relationship between the spatial location of the platen, the matching probability scores of predicted and stencil platen frames are calculated from the row-column dimensions and combined with a correction strategy. Based on maximizing the row and column probability, a one-to-one correspondence is realized between the predicted platen frame in the captured image and stencil platen frame recorded in the disk diagram. Subsequently, the cast-in and cast-out status of the platen is checked, and the staff is notified with an alarm if the status is inconsistent. In the image dataset captured in the field scene of the substation, there were a total of 2685 platens, and the proposed platen checking method successfully matched the predicted platen frames with the stencil platen frames with 100% accuracy. Thus, the method is robust and can be used in the actual platen checking of the substation switchboard cabinets to improve manpower efficiency. Figures and Tables | References | Related Articles | Metrics

Select

Dynamic face reconstruction based on expression-driven tensorial radiance field Mingyang ZHANG, Jincheng WENG, Yang LI 2026, 2026 (4):  63-72.  doi: 10.3969/j.issn.1000-5641.2026.04.007 Abstract ( 0 )   HTML ( 0 )   PDF (1854KB) ( 0 )   Save This paper presents a dynamic face reconstruction method, which takes the tensorial radiance field as the basic network structure of scene expression, and proposes an implicit expression-driven approach based on a multi-layer neural network, extending the expressive capability of the tensorial radiance field to dynamic facial scenes, and optimizes the loss function to further improve the reconstruction effect. Compared with the current method, it can achieve dynamic face representation with higher speed and better quality. Qualitative and quantitative experimental results show that compared with the original methods, our method can save computing resources slightly, improve the training speed by about three times and the inference speed by about ten times, and maintain high-quality image reconstruction results. Figures and Tables | References | Related Articles | Metrics

Select

Road extraction algorithm for remote sensing images guided by deep semantics Xuyang LI, Yan GAO, Hongyan QUAN 2026, 2026 (4):  73-82.  doi: 10.3969/j.issn.1000-5641.2026.04.008 Abstract ( 0 )   HTML ( 0 )   PDF (2392KB) ( 0 )   Save Remote sensing images can provide rich information on road networks, and traditional algorithms can extract details about visible roads from images based on local features. However, such algorithms often encounter notable issues with robustness due to their lack of global semantic information as guidance. To address this issue, we propose an algorithm to extract road areas from remote sensing images based on spatial semantic guidance. The proposed approach combines traditional local information with deep semantic features to learn to initialize a network of roads based on semantic segmentation networks. On this basis, we designed a centerline extraction module (CEM) to extract the centerlines of visible roads as guided by a global semantic model and descriptor. Finally, we obtain a diagram of the structure of the road topology by combining the obtained centerlines with the semantic features using a topology construction algorithm. This ensures the preservation of detailed features by incorporating local details. We evaluated the time performance of the proposed algorithm and its accuracy in extracting roads using publicly available remote sensing image data, and the results confirm that the algorithm performed well and was sufficiently effective to achieve further 3D reconstruction. Visualizations of these findings are also provided. Figures and Tables | References | Related Articles | Metrics

Select

A prompt-driven unified unpaired magnetic resonance image translation method based on generative adversarial networks Qiandi YU, Faming FANG, Guixu ZHANG 2026, 2026 (4):  83-91.  doi: 10.3969/j.issn.1000-5641.2026.04.009 Abstract ( 0 )   HTML ( 0 )   PDF (1160KB) ( 0 )   Save Synthesizing magnetic resonance (MR) images for missing contrast is a critical and challenging task. Existing methods usually construct an implicit contrast-shared space with auxiliary loss and modulate the translation process with one-hot contrast code to achieve multi-contrast translation in a single model. We propose an effective method, ProGAN, which is a prompt-driven unified unpaired MR image translation method based on a generative adversarial network (GAN). ProGAN leverages learnable prompts and text embeddings from contrastive language-image pre-training (CLIP) as external knowledge to generate multi-level dynamic prompts, providing powerful and robust guidance on contrasts. Extensive experiments on two public multi-contrast MRI datasets showed that ProGAN achieved state-of-the-art performance both quantitatively and qualitatively. Figures and Tables | References | Related Articles | Metrics

Select

A unified pansharpening model based on prompt learning Hongfei ZHAO, Tingting WANG, Faming FANG, Guixu ZHANG 2026, 2026 (4):  92-101.  doi: 10.3969/j.issn.1000-5641.2026.04.010 Abstract ( 0 )   HTML ( 0 )   PDF (1316KB) ( 0 )   Save Pansharpening is an image processing technique primarily used to enhance the spatial resolution of remote sensing images by fusing high spatial resolution panchromatic images with low spatial resolution multispectral images. With the development of deep learning, building deep models has gradually become the mainstream method for pansharpening. However, current mainstream technologies are based on training models using datasets from specific satellites, limiting the application of the trained models to fuse images from only those satellite datasets, without covering satellites with limited datasets. Addressing this issue, this paper proposes a unified pansharpening approach that leverages prompt learning technology to train models on datasets from different satellites. The trained unified model can be applied to fuse images from various satellite datasets. Experimental results demonstrate that the model achieves state-of-the-art performance and exhibits good generalization capabilities. Figures and Tables | References | Related Articles | Metrics

Select

Social cognition evaluation framework and question-answering benchmark for large language model Yiming MA, Xin LIN, Qin NI, Yangze YU, Ciping DENG, Liang HE 2026, 2026 (4):  102-111.  doi: 10.3969/j.issn.1000-5641.2026.04.011 Abstract ( 0 )   HTML ( 0 )   PDF (2369KB) ( 0 )   Save Considering the increasing interactions between large language model (LLM) and humans, the assessment of their proficiency in social cognition, a core component of understanding mutual perspectives, is crucial. Drawing from human social cognition theories, this study created an evaluation framework rooted in the theory of mind (ToM), which focuses on the ability to infer the mental states of others. By employing this framework, a social cognition assessment for LLM was developed and presented as a benchmark for multiple-choice questions, inspired by psychological experiments. By applying this benchmark to various commercial models and examining the development of LLM's ToM abilities, our findings indicate that while LLM exhibit nascent ToM skills, there is substantial room for improvement. Furthermore, our study revealed similarities between the development of ToM in LLM and humans. Figures and Tables | References | Related Articles | Metrics

Select

Spatial feature fusion and enhancement for remote sensing road extraction Ziyang XIE, Yan GAO 2026, 2026 (4):  112-122.  doi: 10.3969/j.issn.1000-5641.2026.04.012 Abstract ( 0 )   HTML ( 0 )   PDF (2242KB) ( 0 )   Save Traditional road extraction methods exhibit instability when utilizing data information for processing road images of different types and scales. One major reason for this is that these methods fail to effectively integrate local and global features, demonstrating a weak ability in extracting road details and semantic information. To address this issue, this paper proposes a new road extraction method that employs spatial feature fusion and enhancement strategies. Specifically, the spatial feature fusion module uses graph convolution and local feature extraction to effectively combine global and local features. The spatial feature strength module, on the other hand, applies an attention mechanism to weigh features along both spatial and channel dimensions, enhancing the model’s ability to perceive features, thereby improving its adaptability to road images of different scales. This paper conducted experimental validation of this method on multiple datasets and compared it with existing approaches. The experimental results demonstrate that this method significantly improves performance in road segmentation tasks, offering high robustness and generality, making it suitable for road image datasets of various types and scales. Figures and Tables | References | Related Articles | Metrics

Select

FusionSeg: Topic segmentation model based on semantic feature fusion Zihao GONG, Zhiyun CHEN 2026, 2026 (4):  123-133.  doi: 10.3969/j.issn.1000-5641.2026.04.013 Abstract ( 0 )   HTML ( 0 )   PDF (2995KB) ( 0 )   Save The task of topic segmentation aims to automatically divide text into non-overlapping segments with consistent topics. Traditional topic segmentation models simply map sentences to labels, neglecting the capture of sentence contextual relevance. FusionSeg’s fusion feature encoder dynamically integrates current information with historical information through weights extracted by self-attention mechanisms. Meanwhile, through a combination with fusion contrastive learning, which redefines the positive and negative samples for the topic segmentation task, FusionSeg enhances the role of sentence contextual relevance in the topic segmentation model. In an experimental setup equipped with a single NVIDIA 4090 graphics card, FusionSeg demonstrated a significant improvement in evaluation metric scores, ranging from 1% to 17%, compared to multiple second-best models, such as SeqModel and PEN-NS, across three benchmark datasets: Wiki-727k, En-city, and Wiki-zh. Ablation experiments further demonstrated that FusionSeg’s fusion feature encoder and fusion contrastive learning can capture sentence contextual relevance and optimize topic segmentation results. Figures and Tables | References | Related Articles | Metrics

Select

Transfer learning framework based on linguistic similarity for low-resource neural machine translation Wei YIN, Liyang YU 2026, 2026 (4):  134-142.  doi: 10.3969/j.issn.1000-5641.2026.04.014 Abstract ( 0 )   HTML ( 0 )   PDF (838KB) ( 0 )   Save In contexts where training resources are extremely limited, neural machine translation models based on deep learning often fail to achieve their desired performance. Current methods in transfer learning that leverage similar languages rely on intuition to select analogous data for rudimentary pre-training and potentially fail to pinpoint the most similar languages or fully exploit the advantages of pre-training. Hence, a framework for low-resource machine translation transfer learning based on linguistic similarity is proposed. This framework selects five low-resource languages translated into English as tasks. Initially, six high-resource languages are chosen, and the model is pre-trained on machine translation datasets from these languages to English. Subsequently, employing linguistic similarity metrics, the translation model that is most similar to the target language pair is selected for transfer, ultimately resulting in enhanced model performance via refined fine-tuning strategies. The experimental findings demonstrate that models trained within this framework exhibit superior performance, compared with baseline models, to thereby offer a viable and versatile approach for low-resource machine translation tasks. Figures and Tables | References | Related Articles | Metrics

Select

Automatic generation of use case diagrams from product requirement documents using large language models Zezheng GONG, Fuke SHEN, Tongquan WEI 2026, 2026 (4):  143-153.  doi: 10.3969/j.issn.1000-5641.2026.04.015 Abstract ( 0 )   HTML ( 0 )   PDF (720KB) ( 0 )   Save With the rapid growth of the internet and mobile application industries, agile development has become the preferred approach for product iteration in many organizations. However, as product requirement documents become increasingly complex, the testing process also becomes more complicated, reducing overall development efficiency. To address this challenge, this study investigates the automatic generation of use case diagrams from product requirement documents using large language models. Prompt engineering techniques are employed to optimize the conversion process, enabling the model to accurately understand and extract key information from requirement documents and directly generate use case diagrams that can be readily used by testers. In addition, a comprehensive evaluation framework is developed based on real business scenarios to systematically assess the quality of the generated diagrams. The output format and prompts are further refined through iterative experiments. Experimental results show that the proposed optimization strategy increases the average user adoption rate by 32% and reduces the time required for testers to create test cases by 30%. Figures and Tables | References | Related Articles | Metrics

Select

High-quality codebook priors for magnetic resonance image super-resolution Yuqi LI, Jiaming FAN, Faming FANG, Guixu ZHANG 2026, 2026 (4):  154-164.  doi: 10.3969/j.issn.1000-5641.2026.04.016 Abstract ( 0 )   HTML ( 0 )   PDF (2146KB) ( 0 )   Save Acquiring high-resolution (HR) magnetic resonance (MR) images remains challenging in clinical practice because of limitations in scanning conditions. As an effective post-processing technique for enhancing image resolution, super-resolution (SR) plays an important role in improving the spatial resolution of MR images and supporting clinical diagnosis. Most existing MR image SR methods use deep neural networks to learn a direct mapping from low-resolution (LR) images to HR images. However, LR images lose high-frequency information during the degradation process, making it difficult to reconstruct fine textures and details using only the limited information available in LR images. To improve the accuracy of detail and texture reconstruction, this study introduces codebook priors into MR image SR. The codebook is trained to learn prior information from HR images and is used to guide the reconstruction of LR images. Specifically, we propose a two-stage High-Quality Codebook Prior Network (CPNet). In the first stage, an encoder-decoder network is pretrained on HR images to extract high-quality codebook priors. In the second stage, a Cross-Channel Attention Fusion Module (CCAFM) is developed to incorporate the codebook priors into the SR network. In addition, a novel Gated Convolutional Transformer Block is designed as the basic building block to enhance feature extraction. Experimental results demonstrate the effectiveness of the proposed method and show that it achieves state-of-the-art performance. In particular, the introduction of codebook priors substantially improves the network's ability to restore fine textures and details. Figures and Tables | References | Related Articles | Metrics