2026 Vol.44

Please wait a minute...


For Selected: Download Citations EndNote Ris BibTeX Toggle Thumbnails

Select

Journal of Applied Sciences    2026, 44 (1): 0-0.   Abstract （0）      PDF（pc） （46KB）（3）       Knowledge map    Save Related Articles | Metrics | Comments（0）

Select

Journal of Applied Sciences    2026, 44 (1): 0-0.   Abstract （96）      PDF（pc） （71KB）（33）       Knowledge map    Save Related Articles | Metrics | Comments（0）

Select

Network Community Detection Based on Structure-Enhanced Deep Clustering LI Yongzhen, MA Fuyuan, MA Shixuan, WANG Yuhan, WANG Ying Journal of Applied Sciences    2026, 44 (1): 1-20.   DOI: 10.3969/j.issn.0255-8297.2026.01.001 Abstract （138）      PDF（pc） （3512KB）（92）       Knowledge map    Save Community detection in social networks is important for applications such as information diffusion, recommendation, and advertising. However, existing methods still face challenges in feature fusion, sparse graph modeling, and multi-source information utilization. To address these issues, this paper proposed a structure-enhanced deep clustering (SDC) model for network community detection, which consisted of four key modules. First, the topology enhancement module built an enhanced adjacency matrix by modeling second-order similarity between nodes, which alleviated the problem of missing high-order relations in sparse social networks. Second, the multi-view feature fusion module dynamically fused node attributes and topology features at the node level, and integrated semantic information from both the original and enhanced graphs at the graph level. Third, the multi-source distribution fusion clustering module used learnable weights to integrate clustering information from different feature spaces at the distribution level, balancing local topology and global semantics. Finally, the dual self-supervised module optimized the model through Kullback-Leibler (KL) divergence alignment, node reconstruction, and similarity constraints. Experiments show that compared with mainstream baseline methods, SDC model improves ACC, NMI, ARI, and F1 by an average of 3.80%, 9.09%, 11.21%, and 7.43%, respectively on the three benchmark datasets. Simulations based on Facebook interaction data also demonstrate the ability of SDC model to capture community structure evolution. Reference | Related Articles | Metrics | Comments（0）

Select

Real-Time Travel Pattern Recognition Algorithm Based on Self Adaptive Pooling Enhanced Attention Mechanism LI Yinxiang, DU Wenyuan, XU Zhe, PENG Chen, YAN Jianqiang Journal of Applied Sciences    2026, 44 (1): 21-33.   DOI: 10.3969/j.issn.0255-8297.2026.01.002 Abstract （123）      PDF（pc） （1311KB）（51）       Knowledge map    Save Identifying travel patterns is a crucial task in integrated mobility as a service. To address the limitations of existing traffic travel pattern recognition algorithms, namely insufficient accuracy and real-time application demands, this paper proposed a real-time travel pattern recognition algorithm based on a self adaptive pooling enhanced attention mechanism. Based on the point cloud network, the proposed algorithm efficiently learned current and historical information by incorporating causal convolution and causal pooling, thereby achieving real-time travel pattern recognition. A self adaptive pooling enhanced attention module was further embedded into the framework model to calculate the weight map among different features, thereby enhancing the feature modeling capability. Additionally, the algorithm integrated both motion and geographical features, which effectively improved the recognition accuracy of bus and car travel patterns. Experimental results show that the proposed algorithm achieves superior accuracy. Compared with other one-stage methods, its recognition accuracy is improved by approximately 0.05. Compared with the latest two-stage models such as FPbiLSTM, the parameter count of the proposed algorithm is only 0.167 that of these models, making the proposed approach more lightweight and suitable for deployment on mobile devices. Reference | Related Articles | Metrics | Comments（0）

Select

Complex Logical Query Model Based on Improved Transformer CHEN Yuyin, LI Guanfeng, QIN Jing, XIAO Yuhang Journal of Applied Sciences    2026, 44 (1): 34-49.   DOI: 10.3969/j.issn.0255-8297.2026.01.003 Abstract （117）      PDF（pc） （819KB）（35）       Knowledge map    Save With the widespread application of knowledge graphs in scenarios such as intelligent question answering and recommendation systems, answering complex logical queries on incomplete knowledge graphs has become the focus and difficulty of current research. In view of the fact that ordinary embedding-based methods need to be trained on complex logical queries and cannot be well generalized to query structures outside the distribution, this paper proposed an improved-Transformer-based model DCMHA-MoE that integrated the dynamically composable multi-head attention (DCMHA) mechanism and the mixture-of-experts (MoE) network. This model represented complex query graphs as sequence inputs through triple transformation and bidirectional path encoding technology, and dynamically modeled the structural dependencies and semantic interactions therein, so that complex logical queries can be realized. The DCMHA realized the adaptive combination of attention heads to enhance the semantic expression ability. The MoE network introduced a sparse activation mechanism to improve the adaptability to different query structures and reduce the computational cost. Experiments were conducted on the FB15K-237 and NELL-995 datasets. The results show that compared with the baseline model DiffCLR, the DCMHA-MoE model improves the mean reciprocal rank (MRR) in existential positive first-order logic (EPFO) query $(\wedge, \vee)$ by 10.4% and 7.2%, respectively, which verifies the effectiveness and superiority of DCMHA-MoE in complex logical query tasks. Reference | Related Articles | Metrics | Comments（0）

Select

Uncertain Knowledge Graph Reasoning Model Based on Gaussian Metric Learning ZHANG Yuting, TENG Fei, YE Xiaoqing Journal of Applied Sciences    2026, 44 (1): 50-66.   DOI: 10.3969/j.issn.0255-8297.2026.01.004 Abstract （122）      PDF（pc） （977KB）（31）       Knowledge map    Save Long-tail relations containing only a small number of facts are prevalent in real-world knowledge graphs, and few-shot knowledge graph completion aims to address this data sparsity issue. However, existing approaches often neglect the inherent uncertainty of entities and triples, which limits model reasoning performance under noisy or data-scarce conditions. This paper proposed a covariance-optimized Gaussian metric learning model for uncertain completion (CoGMUC) model to tackle uncertain reasoning in few-shot knowledge graphs. This model represented entities and relations within knowledge graphs as Gaussian distributions, effectively capturing their inherent uncertainty through covariance matrices. It computed semantic similarity with a covariance-aware multi-matching network to complete missing facts and predict confidence levels. Furthermore, a difficult negative sample mining strategy was introduced to enhance the discriminative capability and generalization performance of the model. Experimental results on the public datasets NL27K and CN15K demonstrate that compared with the existing few-shot uncertain knowledge graph completion model based on Gaussian metric learning, CoGMUC improves mean reciprocal rank (MRR) by 21.8% and 2.3% and increases Hits@10 by 9.6% and 21.5%, respectively in the link prediction task. Meanwhile, in the confidence prediction task, the mean squared error (MSE) is reduced by 14.3% and 7.7%, respectively. The findings demonstrate that the CoGMUC model effectively models and leverages uncertainty information, significantly enhancing the performance of few-shot knowledge graph completion. Reference | Related Articles | Metrics | Comments（0）

Select

Construction of Malware Knowledge Graph for Threat Intelligence Analysis XIANG Ga, HU Yan, ZHANG Yangsen, SUN Lu, QI Rui, TAN Zicheng Journal of Applied Sciences    2026, 44 (1): 67-82.   DOI: 10.3969/j.issn.0255-8297.2026.01.005 Abstract （141）      PDF（pc） （852KB）（21）       Knowledge map    Save Threat intelligence analysis is a crucial means to enhance proactive defense capabilities. Research on the construction of malware knowledge graphs holds significant importance for improving malware detection capabilities. In the construction of malware knowledge graphs, the accuracy and completeness of entity and relation extraction still require further improvement. This paper proposed a method for constructing malware knowledge graphs based on a joint extraction model. Firstly, a malware ontology model was proposed for threat intelligence analysis, defining 12 types of relations to standardize the expression of key knowledge about malware. Then, a joint extraction model based on RoBERTa with whole word masking (RoBERTa-Wwm) and pointer annotation was proposed to extract malware entities and their relations, thereby constructing a graph. The experiment demonstrates that the model achieves good performance with an F1 value of up to 0.841. This study is of great significance for the automatic analysis of malware threat intelligence, laying the foundation for improving proactive defense capabilities. Reference | Related Articles | Metrics | Comments（0）

Select

Federated Recommendation Algorithm Integrating Graph Neural Networks and Depth Graph Clustering YI Huawei, SONG Shixi, WANG Yanfei, BAI Siyi Journal of Applied Sciences    2026, 44 (1): 83-96.   DOI: 10.3969/j.issn.0255-8297.2026.01.006 Abstract （122）      PDF（pc） （1065KB）（31）       Knowledge map    Save Federated learning, as the main framework for addressing privacy and security issues in recommendation systems, faces problems such as poor recommendation accuracy, insufficient privacy protection, and excessive communication overhead in practical applications. To address these issues, this paper proposed a federated recommendation algorithm integrating graph neural networks and depth graph clustering. Firstly, a graph neural network was used to capture high-order complex user and item interaction relationships, improving the recommendation accuracy of the recommendation system. Secondly, differential privacy noise was injected into the communication link between the federated learning clients and server to blur the true gradient, thereby enhancing the privacy protection capability of the recommendation system. Finally, clients were clustered by introducing depth graph clustering, and client representatives from each cluster were selected to participate in training. The obtained parameters were shared within the cluster to accelerate the convergence speed of the model and reduce communication overhead under the federated learning framework. The experimental results on real datasets show that the proposed algorithm can enhance privacy protection of the system and reduce communication overhead while improving recommendation accuracy. Reference | Related Articles | Metrics | Comments（0）

Select

Forest Image Dehazing Based on Feature Fusion Attention and Contrastive Learning WU Wenqiang, CHEN Aibin, LI Xiaoyao Journal of Applied Sciences    2026, 44 (1): 97-109.   DOI: 10.3969/j.issn.0255-8297.2026.01.007 Abstract （122）      PDF（pc） （8033KB）（28）       Knowledge map    Save Current algorithms often struggle with incomplete dehazing, loss of details in dark areas, and color distortion in forest foggy images. To address these issues, this paper proposed an adaptive forest image dehazing algorithm based on feature fusion attention and contrastive learning. A multi-scale feature fusion attention mechanism was designed, which dynamically adjusted feature responses by combining channel and spatial attention, thereby enhancing the representation capability of important features. A local contrast regularization module was constructed to enhance the ability of the model to discriminate variations in fog concentration in dark and distant areas. Furthermore, an adaptive color correction module was introduced to mitigate color distortion. Experimental results on both synthetic and real-world forest foggy image datasets demonstrate that the proposed algorithm outperforms existing methods, achieving significant improvements in peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) and a reduction in natural image quality evaluator (NIQE), and exhibiting strong robustness and generalization ability. Reference | Related Articles | Metrics | Comments（0）

Select

Explainable Learning Path Recommendation Based on Dynamic Attention Reinforcement Learning ZHANG Xiaoming, FENG Zejia, WANG Huiyong, ZHANG Xiaojing Journal of Applied Sciences    2026, 44 (1): 110-133.   DOI: 10.3969/j.issn.0255-8297.2026.01.008 Abstract （179）      PDF（pc） （1455KB）（35）       Knowledge map    Save The popularization of large-scale online education has made it difficult for learners to choose courses, and personalized learning path recommendation faces the challenge of relying on single modal data, which leads to the limitation of semantic representation. Moreover, static knowledge maps are difficult to generate dynamic explainable recommendation logic. To address the aforementioned issues, this paper proposed a framework of explainable learning path recommendation based on dynamic attention reinforcement learning (ELPR-DARL). Firstly, a heterogeneous collaborative knowledge graph was constructed, integrating course text, visual content, and knowledge dependencies to enhance cross-modal semantic alignment capabilities. Secondly, a dynamic attention aggregation mechanism for adjacent nodes was designed, which adjusts the weights of entity relationships through a bias correction strategy, and a bidirectional interaction aggregator was utilized to fuse multi-level neighborhood features, enhancing the fine-grained expression ability of knowledge reasoning. Finally, a knowledge graph-aware reinforcement learning strategy was proposed, which explicitly modelled the association between user behavior and knowledge topology based on path connectivity reward functions, generating explainable paths that include global rewards and local attention weights. Experiments based on the MOOC dataset show that this method achieves 22.85%, 33.81%, 52.01%, and 6.34% in NDCG, Recall, HR, and precision metrics, respectively, which is 2.88%, 3.55%, 2.42%, and 3.26% higher than the suboptimal model. User research shows that 80.36% of learners believe that path explanation significantly improves recommendation transparency. This study verifies that the collaborative optimization of a dynamic attention mechanism and reinforcement learning can effectively balance recommendation accuracy and explainability. Reference | Related Articles | Metrics | Comments（0）

Select

Three-Dimensional Fuzzy Clustering Algorithm Integrating Spatial Texture Features JIN Zhengyang, YAN Shaohong, ZHANG Yanbo, YAO Xulong, TAO Zhigang, CHEN Zhiyuan Journal of Applied Sciences    2026, 44 (1): 134-148.   DOI: 10.3969/j.issn.0255-8297.2026.01.009 Abstract （86）      PDF（pc） （7112KB）（19）       Knowledge map    Save The traditional fuzzy C-means (FCM) clustering algorithm is highly sensitive to the initial cluster centers and the noise points. These limitations become more pronounced in complex environments or high-dimensional spaces. To overcome these issues, this study proposed a three-dimensional FCM algorithm integrating spatial texture features. The algorithm was designed to identify regions with noticeable density differences caused by uneven distribution of internal components in the analyzed objects. First, the method extended the two-dimensional gray-level co-occurrence matrix and planar texture feature theory into three-dimensional space to describe spatial texture features. Next, contrast texture features were used to improve the selection of initial cluster centers. Finally, dissimilarity texture features were integrated into the conventional objective function of FCM algorithm to enhance noise resistance. In a simulated experiment on fracture extraction, the proposed algorithm achieved an accuracy of 99.39%, representing a 34% improvement over the traditional FCM algorithm (accuracy of 65.31%). These results confirm the effectiveness of the new algorithm in extracting regions with noticeable density differences inside the analyzed objects. In practical applications, the new algorithm shows superior performance in identifying and extracting human thoracic skeleton. Reference | Related Articles | Metrics | Comments（0）

Select

Multi-granularity Semantic Aspect-Based Sentiment Analysis Model with Fusion of BERT Encoding Layers XU Kai, CHI Mingde, WANG Qi, LI Jianzhou, ZHANG Hui Journal of Applied Sciences    2026, 44 (1): 149-165.   DOI: 10.3969/j.issn.0255-8297.2026.01.010 Abstract （138）      PDF（pc） （1725KB）（57）       Knowledge map    Save Aspect-based sentiment analysis (ABSA) aims to identify the sentiment polarity toward specific aspects within a text. However, existing research still faces multiple challenges: BERT-based approaches suffer from semantic overfitting and insufficient utilization of low-level semantic features; the self-attention mechanism is prone to losing local information; structures with multiple encoding layers and multi-granularity semantics lead to information redundancy. To address these issues, this paper proposed a multi-granularity semantic aspect-based sentiment analysis model with fusion of BERT encoding layers (MSBEL). The model introduced a pyramid attention mechanism to leverage semantic features from various encoding layers, and was combined with low-level encoders to mitigate overfitting. It employed multi-scale gated convolution to enhance its capability in handling local information loss and utilized cosine attention to highlight sentiment features relevant to aspect terms, thereby reducing information redundancy. t-SNE visualization demonstrates that the clustering effect of sentiment representations of MSBEL is superior to that of BERT. MSBEL was compared with mainstream models on multiple benchmark datasets. Compared with LCF-BERT, it achieves F1 improvements of 1.53%, 3.94%, 1.39%, 6.68%, and 5.97% on five datasets. In comparison with SenticGCN, it achieves an average increase of F1 by 0.94% and a maximum increase of 2.12%. Compared with ABSA-DeBERTa, MSBEL increases the F1 by 1.16% on average and achieves a maximum increase of 4.20%. These results validate the effectiveness and superiority of the proposed model for ABSA tasks. Reference | Related Articles | Metrics | Comments（0）

Select

Completion Method for Ship Point Cloud Based on Symmetry Priors ZENG Yinchuan, ZHENG Bo, WANG Xianbao, XIANG Sheng Journal of Applied Sciences    2026, 44 (1): 166-180.   DOI: 10.3969/j.issn.0255-8297.2026.01.011 Abstract （120）      PDF（pc） （2289KB）（77）       Knowledge map    Save Due to the inherent limitations of single-view scanning and the spatial occlusion effects of complex ship hull structures, existing data collection systems commonly face the technical bottleneck of extensive missing data in back-side point clouds. To address this challenge, this paper proposed a ship point cloud completion method based on symmetry priors. This method operated without the need for labeling data and utilized the symmetrical structural characteristics of ships as prior-driven knowledge to effectively complete the back-side point clouds. First, a feature extraction model of the longitudinal centerplane for various types of ship hull was established based on geometric topology analysis of the ships. Then, a symmetry transformation field generation algorithm was proposed to make a mirror completion for the missing point clouds along the longitudinal centerplane of the ship hull, thereby constructing a candidate point cloud set for completion. Finally, an average nearest neighbor quality assessment function between the candidate point clouds and the original point clouds was designed to robustly select the optimal completion result. Experimental results show that the proposed method effectively completes the back-side point clouds of typical ship types, such as sharp-prowed and flat-bottomed ships, without requiring any training samples, and it meets the requirements of real-time data collection scenarios. Reference | Related Articles | Metrics | Comments（0）

Select

Journal of Applied Sciences    2026, 44 (2): 0-0.   Abstract （100）      PDF（pc） （82KB）（49）       Knowledge map    Save Related Articles | Metrics | Comments（0）

Select

Journal of Applied Sciences    2026, 44 (2): 1-1.   Abstract （48）      PDF（pc） （49KB）（11）       Knowledge map    Save Related Articles | Metrics | Comments（0）

Select

A Position Optimization Method for Dual-UAV Double-Assisted Covert Communication LIU Xuedong, TIAN Wen, ZHANG Shuang, DAI Yuewei, SHI Huaifeng Journal of Applied Sciences    2026, 44 (2): 181-197.   DOI: 10.3969/j.issn.0255-8297.2026.02.001 Abstract （141）      PDF（pc） （849KB）（102）       Knowledge map    Save Due to their high mobility and rapid deployment, unmanned aerial vehicles (UAVs) can significantly expand the coverage of covert communication by relay-assisted methods in complex military confrontation areas. However, traditional UAV relay-assisted covert communication suffers from low transmission rates over long distances due to covert constraints. To this end, this paper proposed a position optimization method for dual-UAV double-assisted covert communication. While employing one UAV as a relay to forward covert signals, this method introduces an additional UAV as a friendly jammer to increase the uncertainty at the detector. Specifically, a dual-UAV double-assisted covert communication model was built, the detection error probability of the detector under this model was analyzed, and the covert constraint was derived. Then, the block coordinate descent (BCD)method was employed to jointly optimize the positions of the jamming UAV and the relay UAV under covert and transmission power constraints, aiming to maximize the system’s covert transmission rate. Simulation results demonstrate that the proposed method can significantly improve the covert transmission rate compared to existing methods. Reference | Related Articles | Metrics | Comments（0）

Select

Dynamic Simulation of Heat Treatment Process in Core-Cladding Junction Region of Aluminum-Doped Germanium-Core Optical Fibers ZHONG Shuangqi, DU Yifan, MA Zecheng, XU Sitao, ZHAO Ziwen Journal of Applied Sciences    2026, 44 (2): 198-207.   DOI: 10.3969/j.issn.0255-8297.2026.02.002 Abstract （123）      PDF（pc） （1112KB）（88）       Knowledge map    Save For the heat treatment preparation and modification of pn junction devices in germanium-core fibers, this study selected the core-cladding junction region as the research focus and adopted a typical p-type germanium material with the host element aluminum as the doping element. An aluminum (Al)-doped germanium (Ge)/silicon dioxide (SiO2) interface model was established by using Materials studio software. Dynamic simulations of the heat treatment process for pn junctions in germanium-core fibers were conducted at different temperatures, specifically 500 ℃, 600 ℃, 660 ℃, 700 ℃, 727 ℃, and 827 ℃. The mean square displacement of the dopant atoms Al, diffusion coefficients, and cellular parameter changes of the Al-doped Ge/SiO2 structure were analyzed at different temperatures. It was observed that the diffusion of Al atoms in the structure decreased progressively at 500 ℃, 600 ℃, and 660 ℃, while it increased at 700 ℃, 727 ℃, and 827 ℃. This phenomenon preliminarily indicates the displacements of dopant atoms induced by the temperature increase lead to changes of atomic positions in the crystals and ultimately result in the alteration of the properties of pn junctions within germanium-core fibers. Moreover, the stress variations and stress-strain relationship of the Al-doped Ge/SiO2 structure were investigated. It was found that the internal stress in germanium-core optical fiber is manifested as tensile stress and the trend of tensile stress with temperature is consistent with the changes in structural cell parameters. Additionally, the stress-strain relationship was observed to be proportional at different temperatures. For temperatures of 500 ℃, 600 ℃, 660 ℃, 700 ℃, and 727 ℃, the elastic modulus decreases as the temperature rises; however, when the temperature reached 827 ℃, the elastic modulus increased. These findings provide important insights for optimizing the heat treatment modification of junction-type devices in semiconductor-core fibers. Reference | Related Articles | Metrics | Comments（0）

Select

Routing Optimization for LEO Satellite Networks Based on MAODV algorithm KANG Yujie, SHI Jianfeng, LI Baolong Journal of Applied Sciences    2026, 44 (2): 208-223.   DOI: 10.3969/j.issn.0255-8297.2026.02.003 Abstract （123）      PDF（pc） （969KB）（59）       Knowledge map    Save To address the challenges of frequent link disruptions, long route recovery delays, and poor reliability of traditional single-path routing protocols in low earth orbit satellite networks, this paper constructs a network model that integrates orbital dynamics with link-state awareness and proposes an adaptive routing optimization method based on the multipath ad hoc on-demand distance vector (MAODV) protocol. The proposed method employs an on-demand route discovery mechanism to simultaneously acquire and maintain multiple feasible paths during distance-vector propagation, thereby forming a candidate path set. By further considering link availability fluctuations induced by orbital dynamics, cross-layer parameters—including hop count, link quality, node load, and residual energy—are collected in real time to construct a comprehensive path cost metric, which is used to select both primary and backup paths that are better suited to highly dynamic topologies. A joint simulation environment is established using MATLAB and STK (Satellite Tool Kit), and comparative experiments with representative routing protocols are conducted to evaluate the performance of the proposed method. Simulation results demonstrate that, under identical constellation and traffic conditions, the proposed method achieves average throughput improvements of approximately 4.29%, 4.77%, and 3.82% over the ad hoc on-demand distance vector (AODV), dynamic source routing (DSR), and optimized link state routing (OLSR) protocols, respectively. Moreover, the average end-to-end delay is significantly reduced compared with these baseline protocols, thereby confirming the effectiveness of the proposed approach in dynamic LEO satellite network scenarios. Reference | Related Articles | Metrics | Comments（0）

Select

Reversible Data Hiding Algorithm Using Structural Similarity Index Measure GUO Kexin, XIANG Shijun Journal of Applied Sciences    2026, 44 (2): 224-233.   DOI: 10.3969/j.issn.0255-8297.2026.02.004 Abstract （105）      PDF（pc） （2491KB）（67）       Knowledge map    Save The distortions in reversible data hiding (RDH) include pixel distortion and structural distortion. With the high sensitivity of the human visual system to structural distortions in images considered, this paper adopted the structural similarity index measure (SSIM) as the evaluation metric for RDH. First, by analyzing the theoretical gain relationship between peak signal-to-noise ratio (PSNR) and SSIM, the dynamic and simultaneous evaluation of the two metrics was realized. Subsequently, a high-SSIM RDH method based on texture region prioritization was proposed. Image preprocessing was performed by dividing the carrier image into four independent pixel sets, and texture regions were accurately located. Data was then embedded in descending order of background complexity. Experimental results show that the strategy of prioritizing data embedding in texture regions reduces the structural distortion of images. At the same embedding rate, the SSIM value is improved, and the visual quality of images is enhanced. Reference | Related Articles | Metrics | Comments（0）

Select

Multi-view Joint Adjustment Registration Method for Images and Point Clouds with Line Feature Constraints CHEN Lu, WANG Anni, LAN Ziyu, XU Hui, ZHANG Penglin Journal of Applied Sciences    2026, 44 (2): 234-249.   DOI: 10.3969/j.issn.0255-8297.2026.02.005 Abstract （115）      PDF（pc） （44490KB）（21）       Knowledge map    Save The registration of images and laser point clouds is a key technology for 3D scene reconstruction, providing critical scientific support for origin analysis of explosion accidents and holding broad application prospects in fields such as autonomous driving and disaster origin tracing. However, the spatial scale and geometric feature differences between 2D and 3D data collected by heterogeneous sensors pose challenges to the refined registration of image and point cloud modalities. To this end, this paper proposed a multi-view joint adjustment registration method based on line feature constraints. Firstly, line features were extracted from both images and point cloud data for coarse registration. Then, based on the standard perspective-n-point (PnP) model, constraints of directional consistency and orthogonal consistency of line features were introduced. By aiming to minimize the error function across multiple views, the transformation parameter solution was transformed into a nonlinear least squares problem for iterative optimization, ultimately achieving the accurate registration of images and point clouds. This process does not require 2D-3D projection transformations or scale conversions, thus preventing the introduction of projection errors. Comparative experiments show that the proposed multi-view joint adjustment method with line feature constraints can significantly improve the registration accuracy of image and point cloud modalities, reducing registration errors by over 60% compared to the single-view standard PnP model. Reference | Related Articles | Metrics | Comments（0）

Select

Real-Time Semantic Segmentation Based on Composite Three-Branch and Deep Feature Encoding LEI Xiaochun, PAN Yiwei, ZHANG Yongya, JIANG Zetao, LI Mengtong Journal of Applied Sciences    2026, 44 (2): 250-265.   DOI: 10.3969/j.issn.0255-8297.2026.02.006 Abstract （116）      PDF（pc） （15531KB）（26）       Knowledge map    Save To address the problems of small-object segmentation errors and holes in large-object segmentation results in scenes with significant differences in object sizes in real-time semantic segmentation, this paper proposed a real-time semantic segmentation algorithm based on a composite three-branch and deep feature encoding, consisting of a composite three-branch module (CTBM), a deep feature encoding module (DFEM), and a dual-branch multi-layer perceptron (DBMLP). The CTBM used a dual-layer multi-scale feature extraction and fusion strategy to comprehensively extract information from different perspectives, enabling the model to perceive the global relationships between features better and reduce the holes in the large-object segmentation results. The DFEM enhanced the model’ s ability to express deep features through encoding methods, better perceived the semantic information of small objects, and improved the segmentation accuracy of small objects. The DBMLP effectively integrated multi-scale semantic information by utilizing both global and local features, resulting in smoother edges and more accurate contours in segmentation results. Evaluation results on the Cityscapes and ADE20K datasets have shown that the algorithm not only meets real-time speed requirements but also achieves mIoU of 74.2% and 40.4% at 42.6 FPS and 45.3 FPS, respectively, significantly outperforming other real-time semantic segmentation algorithms. Reference | Related Articles | Metrics | Comments（0）

Select

Human Action Recognition in Infrared Images Based on Median-Guided Multi-scale Feature Fusion YUAN Shuai, YU Lei, YAO Tian, XIONG Bangshu Journal of Applied Sciences    2026, 44 (2): 266-281.   DOI: 10.3969/j.issn.0255-8297.2026.02.007 Abstract （106）      PDF（pc） （4501KB）（21）       Knowledge map    Save Conventional deep learning models exhibit limited recognition performance in infrared images, primarily because the lack of discriminative features makes them difficult to effectively distinguish similar behaviors. To solve this problem, a novel infrared image-oriented human action recognition method based on median-guided multi-scale feature fusion was proposed. First, an information modeling mechanism that integrated median-enhanced attention and multi-scale feature comparison was constructed. This mechanism finely modeled the differences between feature hierarchies, guiding the network to focus on the key feature regions that distinguished different action categories, therefore breaking through the limitation of traditional methods that relied on global features for classification. Second, a median-enhanced spatial and channel attention module was designed, which solved the problem that traditional Siamese networks were difficult to accurately focus on the key regions of the human body in infrared action images due to the lack of explicit positional information in deep features. Finally, a multi-scale feature fusion module was proposed, which could effectively fuse multi-scale features, enhance the expression ability of action details and structural information in infrared images, strengthen the model’s ability to capture subtle action changes, and reduce the misjudgment rate caused by information loss and background interference. Experimental results show that the recognition accuracy of the proposed method is superior to that of existing mainstream methods in multiple datasets such as infrared splicing, PUB, and VAIS, which fully demonstrates the effectiveness and advancement of this method. Reference | Related Articles | Metrics | Comments（0）

Select

Multi-Dimensional Parallel Blockchain Suitable for IoT Environment SI Xueming, MA Shuosen, YAO Zhongyuan Journal of Applied Sciences    2026, 44 (2): 282-296.   DOI: 10.3969/j.issn.0255-8297.2026.02.008 Abstract （115）      PDF（pc） （677KB）（29）       Knowledge map    Save As the application of blockchain in the internet of things scenario continues to gain popularity, it becomes increasingly difficult to meet the real-time transaction processing requirements of IoT devices, and the issue of low blockchain throughput becomes increasingly prominent. Existing solutions often rely on technologies such as blockchain sharding, block pipeline, and concurrency control to enhance throughput, but these solutions do not significantly improve throughput in some conflict scenarios. This paper analyzed the characteristics of smart contract transactions of hotspots and the applicable scenarios of various concurrency control technologies. Relying on blockchain sharding technology, it constructed shards for smart contracts of hotspots with high conflict rates and shards for ordinary contracts with low conflict rates. It adapted corresponding concurrency control technologies for the two types of shards, fully leveraging the advantages of each technology to ultimately achieve efficient multi-dimensional parallelism between shards and transactions. Additionally, this paper proposed a formula for evaluating the performance of trusted sensors, assigning adaptive data transmission tasks based on different sensor capabilities, thereby enhancing overall data transmission efficiency. Experimental results have demonstrated that the improvement of throughput of the proposed solution is significant. The solution improves performance more efficiently compared to other single-dimensional parallelism solutions. Reference | Related Articles | Metrics | Comments（0）

Select

A Fast Algorithm for Matrix Multiplication Based on “Regularization-Filtering-Resampling” DING Guangtai, LIU Tong, ZHI Xiaoli, WU Pin, TONG Weiqin Journal of Applied Sciences    2026, 44 (2): 297-315.   DOI: 10.3969/j.issn.0255-8297.2026.02.009 Abstract （128）      PDF（pc） （736KB）（48）       Knowledge map    Save Focusing on the trade-off in terms of speed, accuracy and efficiency between the exact and approximate algorithms for large-scale matrix multiplication, this paper proposed a fast algorithm for dense matrix multiplication employing regularization, filtering, and resampling techniques. Based on the sampling theorem, a regularization relationship between the matrix and its corresponding analog function was established, and then filtering and resampling stages were introduced to achieve the trade-off mechanism between the exact algorithm and the approximate algorithm. In pursuit of higher algorithmic efficiency, the applicable scope and conditions of the algorithm were investigated, especially the relationship between the algorithm accuracy and the statistical characteristics of the matrix data. Data experiments were conducted using matrices generated by methods such as independent and identically distributed random number generators. The results indicate that the algorithm achieves its trade-off objectives. Reference | Related Articles | Metrics | Comments（0）

Select

Breeding Score Algorithm for Succulent Leaves based on Multi-channel Feature Fusion of Area, Contour, and Color ZHANG Lei, LI Taoyuan, LI Mingfei, DENG Haimin, XIE Cheng Journal of Applied Sciences    2026, 44 (2): 316-329.   DOI: 10.3969/j.issn.0255-8297.2026.02.010 Abstract （125）      PDF（pc） （7126KB）（67）       Knowledge map    Save To enhance the accuracy of image scoring and consistency in screening for the propagation potential of succulent leaves, we propose a breeding scoring algorithm based on multi-channel feature fusion of “surface-outline-color”. This algorithm first employs an image segmentation model to automatically detect and locate areas of succulent leaves with breeding potential within the original image, completing preliminary screening. Subsequently, it extracts leaf area, contour, color, and high-level semantic features to construct a unified multidimensional fusion vector, which is then input into a scoring model to generate breeding scores. Among these, the area feature reflects developmental fullness, the contour feature indicates morphological regularity, and the color feature measures health status. Meanwhile, semantic features extracted by a pre-trained visual model supplement deeper growth patterns and semantic associations that traditional metrics struggle to capture, further enhancing the model’s discriminative power and generalization capability. Experimental results demonstrate that this method significantly improves the efficiency and accuracy of leaf screening. Compared to traditional methods, the Pearson correlation coefficient and area under the curve improved by an average of 0.093 8 and 0.065 3, respectively, across multiple succulent datasets. The mean squared logarithmic error decreased by approximately 0.012 1, demonstrating enhanced accuracy and robustness. This effectively supports the intelligent breeding of succulent plants. Reference | Related Articles | Metrics | Comments（0）

Select

Parallel TimesNet-Informer Model for Process Quality Prediction Using STL Decomposition and Crested Porcupine Optimizer LIANG Xinyan, SUN Jingyun, CAI Guojing, CHEN Hailong Journal of Applied Sciences    2026, 44 (2): 330-344.   DOI: 10.3969/j.issn.0255-8297.2026.02.011 Abstract （132）      PDF（pc） （1752KB）（48）       Knowledge map    Save In process manufacturing, quality prediction is particularly challenging due to high levels of noise, complex nonlinear dynamics, and multiscale temporal dependencies in process data. To address these issues, this paper proposes a parallel TimesNet-Informer deep learning prediction model that integrates seasonal-trend decomposition using loess (STL) with the crested porcupine optimizer (CPO). First, the STL method is employed to decompose the original time series into trend, seasonal, and residual components, thereby extracting multiscale temporal features. Second, the CPO algorithm is utilized to auto-matically optimize decomposition parameters and model hyperparameters in a data-driven manner. A parallel architecture is designed that combines the strength of TimesNet in capturing periodic and local features with Informer’s superior capability in modeling long-sequence dependencies, enabling accurate fitting and prediction of complex process quality. The model is validated on real-world data from a process manufacturing production line. Experimental results demonstrate that the proposed model outperforms other comparative models across all evaluation metrics, achieving a prediction accuracy (R2) of 0.979 8. This provides an effective solution for accurate quality prediction in process manufacturing. Reference | Related Articles | Metrics | Comments（0）

Select

Journal of Applied Sciences    2026, 44 (3): 0-0.   Abstract （19）      PDF（pc） （77KB）（25）       Knowledge map    Save Related Articles | Metrics | Comments（0）

Select

Journal of Applied Sciences    2026, 44 (3): 1-0.   Abstract （14）      PDF（pc） （47KB）（6）       Knowledge map    Save Related Articles | Metrics | Comments（0）

Select

Compressed Sensing-Based Approach for Reducing Wear in AFM Fiber Probes YE Shuai, SHANG Yana, CHEN Na, LIU Shupeng, LIU Yong Journal of Applied Sciences    2026, 44 (3): 345-357.   DOI: 10.3969/j.issn.0255-8297.2026.03.001 Abstract （31）      PDF（pc） （8545KB）（15）       Knowledge map    Save To address the issue of tip wear in atomic force microscope(AFM) fiber probe scanning imaging, an undersampling scanning method based on compressed sensing(CS)was proposed, which reduced the number of contacts between the fiber probe and the sample and effectively prolonged the probe lifetime. Complete AFM images were reconstructed using the CS algorithm. A convolutional neural network(CNN) was introduced to optimize the quality of CS-reconstructed images and mitigate the degradation of image quality caused by undersampling. Fiber probe wear was therefore reduced while maintaining high-quality imaging. In the experiments, when the scanning frequency was 0.3 Hz and the number of scanning points was reduced from 200 × 200 to 100 × 100, the tip wear was reduced from 78 nm to 13 nm, and the imaging time was reduced to one quarter of the original imaging time. After optimization using the CNN, the AFM images had a peak signal-to-noise ratio(PSNR) of 30.12 dB and a structural similarity(SSIM) of 0.96. The results demonstrate that low-wear, fast, and high-quality AFM imaging with fiber probes can be achieved. Reference | Related Articles | Metrics | Comments（0）

Select

Uncertainty Modeling-Driven LiDAR-Vision Adaptive Fusion for Dynamic Obstacle Detection ZHU Lei, ZHONG Ruofei, YUAN Xinze, FAN Hongchao, SUN Zhenxing Journal of Applied Sciences    2026, 44 (3): 358-376.   DOI: 10.3969/j.issn.0255-8297.2026.03.002 Abstract （38）      PDF（pc） （15908KB）（15）       Knowledge map    Save Accurate and reliable dynamic obstacle perception is a crucial prerequisite for the safe navigation of autonomous mapping UAVs in confined and spatially constrained environments. However, existing LiDAR-vision fusion methods often struggle to cope with significant variations in sensor reliability under degraded conditions, such as poor illumination, reflection interference, or motion blur. To address these challenges, this paper proposed an uncertainty-modeling-driven LiDAR-vision fusion framework for dynamic obstacle detection, which adaptively adjusted sensor contributions by explicitly modeling sensor observation uncertainty. Based on probabilistic models, the framework performed real-time uncertainty quantification for both LiDAR point clouds and RGB images and introduced an adaptive sensor reliability score(ASRS) mechanism to guide fusion decisions and subsequent object tracking. Experiments were conducted on a self-constructed multi-condition dataset, and the results show that the proposed method improves the F1-score by approximately 15%-20% compared with existing methods in challenging scenarios involving low illumination, glass reflections, and motion blur. Furthermore, it maintains real-time processing performance at approximately 25 Hz on embedded platforms, validating the method's robustness and engineering feasibility in complex degraded environments. Reference | Related Articles | Metrics | Comments（0）

Select

Method for Extracting Chinese-Burmese Parallel Sentence Pairs Based on Language Feature Enhancement ZHAO Zixiao, WANG Hao, SHEN Tao, JIANG Shuting, ZHANG Siqi, LAI Hua, HUANG Yuxin, YU Zhengtao Journal of Applied Sciences    2026, 44 (3): 377-389.   DOI: 10.3969/j.issn.0255-8297.2026.03.003 Abstract （39）      PDF（pc） （280KB）（16）       Knowledge map    Save To address the scarcity of labeled resources and the limited representational capacity of models in extracting parallel sentence pairs in low-resource languages, this paper proposed a language-feature-enhanced method for Chinese-Burmese parallel sentence pair extraction. The method was optimized from three aspects: data augmentation, model architecture, and training mechanism. First, a Chinese-Burmese dual encoder based on a Siamese network was constructed to build a cross-lingual semantic representation space.Second, an information-content evaluation mechanism based on the L2 norm of word vectors was introduced to replace high-information features and perform sample augmentation,thus alleviating the data sparsity problem under low-resource conditions. Finally, positive and negative samples were constructed and dynamically modeled through contrastive learning to optimize sample boundaries and achieve more accurate Chinese-Burmese semantic alignment. Experimental results show that the proposed method achieves an F1 score of 95.03% on the Chinese-Burmese parallel sentence pair extraction task, outperforming the baseline model. In addition, this paper constructs a high-quality general-domain ChineseBurmese dataset containing 5 × 105 sentence pairs, providing data support for research on low-resource languages. Reference | Related Articles | Metrics | Comments（0）

Select

Combustion Chamber Ignition Prediction Algorithm Based on AttResVGG Model LIAO Qing, CHEN Hongyou, LIU Chongyang, QU Lingfeng, DUAN Jiping, XIA Ping, TIAN Baodan, FAN Yong Journal of Applied Sciences    2026, 44 (3): 390-408.   DOI: 10.3969/j.issn.0255-8297.2026.03.004 Abstract （38）      PDF（pc） （1126KB）（10）       Knowledge map    Save To address the challenges posed by complex operating conditions, limited training samples, data imbalance, and the inability of conventional deep learning models to meet the requirements of ignition prediction for aero-engine combustion chambers, a network model integrating residual connections and a self-attention mechanism, called the attention residual visual geometry group network(AttResVGG), was proposed. The model used a multi-head self-attention mechanism to capture dependencies among operating condition parameters and establish a mapping between these parameters and ignition status. To address insufficient data size and class imbalance, a physically constrained data augmentation strategy was designed to synthesize new operating condition samples while maintaining key physical parameter relationships, such as the fuel-air ratio and temperature-pressure ratio.In addition, an automated machine learning algorithm based on Bayesian optimization was designed to optimize model hyperparameters, further enhancing the model's predictive performance. To validate the effectiveness of this model, experiments on two datasets show that the accuracy of the AttResVGG model reaches 97.67% and 90.48%, and the Kappa coefficients reach 0.950 5 and 0.834 6, respectively, which are better than those of the compared models. Reference | Related Articles | Metrics | Comments（0）

Select

Analysis on the External Action Intensity and Spatial Network Relationship of Urban Green Competitiveness: A Case Study of Liaoning Province GUO Lina, YANG Yueheng, ZHAO Yanxia Journal of Applied Sciences    2026, 44 (3): 409-421.   DOI: 10.3969/j.issn.0255-8297.2026.03.005 Abstract （24）      PDF（pc） （1487KB）（6）       Knowledge map    Save With the continued expansion of cities and the rapid growth of urban agglomerations, problems such as unbalanced spatial development and uneven resource allocation have emerged. Optimizing urban spatial structure has become a key issue. This study used principal component analysis (PCA) to measure the quality level of urban green competitiveness in Liaoning Province from 2015 to 2020 and applied a modified gravity model to analyze external action intensity and spatial network relationships. The results are as follows. 1) The overall quality level of green competitiveness of urban agglomerations in Liaoning Province is constantly improving, but the pattern of imbalance between cities with strong and weak competitiveness has not been significantly improved. 2) The external action intensity of cities in Liaoning Province has improved to a certain extent, but differences remain among cities, especially in Shenyang and Benxi, which have values far higher than those of other cities. 3) The network structure of the overall interaction intensity among urban agglomerations in Liaoning Province is relatively dense, with frequent and dynamic interactions among cities. Reference | Related Articles | Metrics | Comments（0）

Select

A Low-Light Few-Shot Object Detection Method Based on Feature Optimization JIANG Zetao, JIN Xin, LENG Lu, ZHU Wencai Journal of Applied Sciences    2026, 44 (3): 422-436.   DOI: 10.3969/j.issn.0255-8297.2026.03.006 Abstract （34）      PDF（pc） （4132KB）（14）       Knowledge map    Save To address the scarcity of samples for low-light object detection in certain environments, this paper proposed a low-light few-shot object detection method based on feature optimization. The method designed a denoising Wasserstein autoencoder(DNWAE)module and an adaptive variational feature aggregation(AVFA) module to address the problem of weak image feature information under low-light conditions and enhance important features. To reduce object classification confusion caused by limited training samples in few-shot learning, the paper designed a category information guided detection head(CIGDH) module to improve detection accuracy. Experimental results show that, compared with the selected mainstream few-shot object detection algorithms, this method achieves an average improvement of 9.3%–19.2% in detection accuracy after being trained on low-light datasets. Moreover, after being trained on normal-light datasets, this method achieves an average improvement of 3.0% in detection accuracy compared with the current state-of-the-art algorithm. The proposed algorithm is meaningful and has good application value for few-shot object detection under low-light conditions. Reference | Related Articles | Metrics | Comments（0）

Select

Dual Attention-Incorporated Lightweight U-shaped Network for Lung Lesion Image Segmentation HE Xiaochen, DING Derui, LI Ming, WANG Fei, WANG Bo Journal of Applied Sciences    2026, 44 (3): 437-451.   DOI: 10.3969/j.issn.0255-8297.2026.03.007 Abstract （23）      PDF（pc） （1936KB）（11）       Knowledge map    Save To address the problems of low contrast, fuzzy texture details, and inadequate edge feature extraction in lung lesion images, this paper proposed a novel lightweight U-shaped network incorporating dual attention, termed DALU-Net. First, an attentionbased two-branch fusion module was designed. During encoding, the two branches focused on global and local information, respectively, to capture global localization information and lesion edge features. Then, a parallel texture enhancement module was introduced,and a statistical feature histogram was obtained using a quantization counting operator to enhance the texture features extracted by the shallow network and address the challenge of low contrast. Finally, a reverse-attention dual-interference refinement module was developed to enable the network to focus on and process mis-segmentation information during the decoding stage, thus simultaneously eliminating false-positive and false-negative features in the reconstructed image. The effectiveness of the network was verified on two lung lesion datasets: COVID-19 CT scan and MS COVID-19. Compared with existing networks,the proposed network achieves the best results on all five evaluation metrics, with a Dice score that is 1.42% higher than that of the second-best model, UNeXt, while also using fewer parameters. Reference | Related Articles | Metrics | Comments（0）

Select

Light Field Image Compression Based on Implicit Disparity Compensation LU Yongjie, AN Ping, HUANG Xinpeng, YANG Chao Journal of Applied Sciences    2026, 44 (3): 452-464.   DOI: 10.3969/j.issn.0255-8297.2026.03.008 Abstract （20）      PDF（pc） （12731KB）（11）       Knowledge map    Save Light field(LF) imaging captures both the positional and angular information of light rays, leading to a significant increase in LF data volume due to its high-dimensional characteristics. As a result, efficient LF compression techniques have become an important research focus in this field. In recent years, researchers have proposed various deep learningbased methods for LF compression. However, these methods often struggle to achieve endto-end joint optimization and require the explicit transmission of disparity or geometric information, which significantly increases the complexity of the coding scheme. To address this issue, this paper proposed a novel end-to-end LF compression model. The model utilized disparity relationships among LF views and used a deformable attention mechanism for disparity compensation, enabling effective LF compression by encoding and decoding disparity features and residuals. Experimental results show that the proposed method outperforms other LF compression methods in rate-distortion performance and achieves state-of-the-art performance in mid-to-high bitrate coding. Reference | Related Articles | Metrics | Comments（0）

Select

Split-Chain Trading System Design for Multi-source Heterogeneous Data Assets LI Shouwei, JIANG Yimin, ZHANG Jiazheng Journal of Applied Sciences    2026, 44 (3): 465-485.   DOI: 10.3969/j.issn.0255-8297.2026.03.009 Abstract （23）      PDF（pc） （580KB）（5）       Knowledge map    Save This paper proposed a split-chain trading system combining a consortium chain and a sidechain for multi-source heterogeneous data assets to address challenges in data asset trading. These challenges include fragmented enterprise and individual trading environments, high risks of privacy leakage, and insufficient system throughput. The transaction characteristics of multi-source heterogeneous data assets were analyzed, and a user-type-aware automatic split-chain routing mechanism was designed. The consortium chain ensures strong-consistency auditing for enterprise data, while the sidechain enables high-concurrency processing of personal data. Trusted computing technology was also introduced to construct a hardware-level privacy protection solution, allowing data to be used without being exposed or copied. The experimental results show that the split-chain architecture increases the peak system throughput to 2 400 TPS, representing an improvement of approximately 118% compared with traditional single-chain systems. Meanwhile, it maintains transaction latency below 2 s under high-concurrency conditions, with the added security overhead remaining at the millisecond level. This system effectively balances security and transaction efficiency while enabling integrated trading of heterogeneous data.It also integrates the trading of enterprise and personal data assets into a standardized transaction system while protecting data privacy, thus providing an engineering solution for data circulation in the digital economy. Reference | Related Articles | Metrics | Comments（0）

Select

Cross-View Group Recommendation Algorithm Driven by Hypergraph Neural Networks GUO Yan, WANG Haoran, HOU Songsong, DUAN Xuliang, MU Jiong Journal of Applied Sciences    2026, 44 (3): 486-502.   DOI: 10.3969/j.issn.0255-8297.2026.03.010 Abstract （37）      PDF（pc） （2122KB）（10）       Knowledge map    Save In recent years, with the development of information technology and the widespread popularity of online social networking, traditional recommender systems have gradually exposed limitations in characterizing complex user needs, thus drawing increasing attention to group recommendation systems. However, most existing group recommendation methods rely on simple aggregation of individual members' preferences, making it difficult to capture implicit consensus in group behavior. To address this problem, this paper proposed IConcen, a hypergraph neural network-based group recommendation model that integrated multi-view information. The model captured group interactions from three complementary perspectives: member level, item level, and group level. It also introduced an adaptive fusion module that dynamically balances the weights of these views to generate highly expressive fused features, effectively improving recommendation performance.Experimental results on the Mafengwo and CAMRa2011 datasets show that the proposed method outperforms mainstream models such as HCR in both group recommendation and user recommendation tasks. Specifically, on the Mafengwo dataset, the proposed method improves HR@5 and NDCG@5 by more than 24.3% and 28.2%, respectively, compared with the S2-HHGR, Agree, and HCR models, verifying its effectiveness and superiority. Reference | Related Articles | Metrics | Comments（0）

Select

Multi-modal Rumor Detection Method Fusing Image and Text Features GAO Guangliang, LIANG Weichao, ZHU Tao, HONG Lei, XIA Lingling Journal of Applied Sciences    2026, 44 (3): 503-514.   DOI: 10.3969/j.issn.0255-8297.2026.03.011 Abstract （23）      PDF（pc） （7701KB）（8）       Knowledge map    Save To further improve the effectiveness and stability of rumor detection, a multimodal rumor detection method was proposed that integrated image and text features with adaptive noise resistance and semantic restoration. First, a three-stage processing pipeline consisting of pretrained encoding, dynamic pooling, and multi-head enhancement was designed to encode rumor-related texts and comments into semantic vectors. Then,two parallel modules were constructed: one for visual feature extraction and the other for optical character feature extraction. These modules encoded rumor images and explicit text within them into complementary enhanced image vectors. Finally, adaptive gating and dynamic cross-attention mechanisms were used to filter noise and enhance semantics,achieving local alignment and global integration of image and text information. Experimental results show that, compared with baseline algorithms, the proposed method can effectively capture deep correlations between image and text information and improve credibility and practicality of rumor detection results. Reference | Related Articles | Metrics | Comments（0）