Applied Intelligence最新文献

{"title":"Correction: Real-time rehabilitation assessment and corrective guidance driven by dual regulation pose analysis","authors":"Yi Qiao, Zilong Wang","doi":"10.1007/s10489-026-07266-2","DOIUrl":"10.1007/s10489-026-07266-2","url":null,"abstract":"","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 7","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147829831","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep learning models for digital medical imaging: a survey","authors":"Bing Liu,&nbsp;Xueju Wang,&nbsp;Yujia Cong,&nbsp;Lele Cong,&nbsp;Xianling Cong,&nbsp;Shisong Tang,&nbsp;Hechang Chen","doi":"10.1007/s10489-026-07211-3","DOIUrl":"10.1007/s10489-026-07211-3","url":null,"abstract":"<div><p>Research on the application of deep learning in the field of medical imaging has achieved significant progress, demonstrating exceptional performance, particularly in disease diagnosis and classification. Compared to clinicians, deep learning models can provide diagnostic results more rapidly for common diseases while maintaining stable accuracy, which holds positive implications for the smooth execution of clinical work. However, numerous challenges remain for the widespread application of deep learning models in clinical practice. Current research primarily focuses on the development of models, with insufficient attention paid to implementation issues in real-world applications. To address this research gap, this paper, based on the “input-model-output\" framework of deep learning, delves into the challenges and corresponding solutions that models may encounter in practical applications from three dimensions. At the input level, we analyze data volume and feature selection, the complexity of model preprocessing steps, and the challenges of data fusion; at the model level, we primarily explore issues such as model generalizability, robustness, and interpretability; at the output level, we summarize three key modules: disease segmentation, diagnosis, and classification. Furthermore, this paper suggests potential future development directions, including the integration of interdisciplinary expert knowledge, the construction of novel medical models, the development of models for non-solid tumors, and the establishment of integrated multi-data web platforms. The research in this paper not only fills the final gap in the promotion and application of deep learning models in the medical field, making a positive contribution to their successful integration into clinical practice, but also provides clinical staff with a deeper perspective for understanding deep learning models, thereby facilitating the integration and collaborative development of the medical and artificial intelligence fields.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 7","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147830063","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An efficient channel attention-enhanced time-frequency masked autoencoder-based time series anomaly detection method","authors":"Yashuang Mu,&nbsp;Zeng Kong,&nbsp;Maoqing Zhang,&nbsp;Hongyue Guo,&nbsp;Lidong Wang","doi":"10.1007/s10489-026-07248-4","DOIUrl":"10.1007/s10489-026-07248-4","url":null,"abstract":"<div><p>Anomaly detection techniques can effectively identify observations that deviate from normal behavior patterns in time series. The current reconstruction-based unsupervised anomaly detection methods still face some challenges, such as key feature extraction and data distribution drift. To address these issues, we propose an Efficient Channel Attention-based Time-frequency Masked AutoEncoder model (ECAT-MAE) for anomaly detection in time series. The proposed method innovatively integrates time-frequency analysis and channel attention mechanisms through comparative criteria within the dual-branch transformer autoencoder architecture. First, an efficient channel attention module is developed via lightweight convolutional operations to dynamically calibrate channel weights and suppress redundant features. Subsequently, a weighted time-frequency masking strategy guided by the recalibrated channel weights is employed to facilitate the learning of unbiased representations of normal patterns, thereby enhancing sensitivity to critical time-frequency features. Finally, we employ a time-frequency consistency contrastive loss function combined with an adversarial training strategy to prevent overfitting and thereby alleviate distribution shift issues. The experimental results indicate that the proposed anomaly detection method achieves higher detection accuracy across several benchmark datasets.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 7","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147829334","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Improve domain generalization through curve enhancement and style structure decoupling for medical image segmentation","authors":"Ruibing Fu,&nbsp;Haoran Zhou,&nbsp;Ziyuan Zhang,&nbsp;Runze Lu,&nbsp;Guangyao Li","doi":"10.1007/s10489-026-07263-5","DOIUrl":"10.1007/s10489-026-07263-5","url":null,"abstract":"<div><p>Medical imaging data (e.g., CT and MRI) often originate from heterogeneous sources, leading to appearance discrepancies and consequently limited generalization when training data are scarce. Prior unsupervised domain adaptation (UDA) and multi-source domain generalization (MSDG) methods mitigate distribution shifts but require access to multiple domains during training. Single-source domain generalization (SSDG) offers a practical alternative. However, existing SSDG methods typically depend on global random image augmentations or implicit feature decoupling, which often result in residual style leakage and unstable cross-domain performance. To address this challenge, we propose GLASSD (Global-Local Augmentation with Style-Structure Decoupling), a novel SSDG framework that explicitly couples learnable data-level augmentation with feature-level decoupling. Specifically, a B-spline-based augmentation module samples knot vectors to parameterize deformation fields, enabling more diverse geometric variations than conventional affine or elastic transforms and thereby improving robustness to domain shifts. Meanwhile, a style-structure decoupling module disentangles style information from structural features, mitigating domain-specific style biases in downstream segmentation. We evaluate GLASSD on two challenging multimodal datasets and benchmark its performance against state-of-the-art domain generalization methods. Extensive experiments demonstrate that GLASSD achieves superior cross-domain performance. Our code is publicly available at https://github.com/bri-bing/GLASSD-main.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 7","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147829332","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Community partitioning and node belonging prediction method based on neighborhood attribute reduction and fuzzy rough set with application in clinical decision making","authors":"Haoran Sun,&nbsp;Xiangtang Chen,&nbsp;Bingzhen Sun,&nbsp;Xixuan Zhao,&nbsp;Xiaoli Chu,&nbsp;Jianxiong Cai","doi":"10.1007/s10489-026-07258-2","DOIUrl":"10.1007/s10489-026-07258-2","url":null,"abstract":"<div><p>The combined diagnosis data of Traditional Chinese and Western medicine is characterized by intricate structures and high dimensionality. These complexities present significant challenges for clinical decision-making. Community partitioning offers a powerful approach to uncover hidden relationships within such medical data. It provides theoretical support for discovering potential comorbidities and formulating personalized treatment plans. To address attribute redundancy in medical datasets, this study proposes a novel community partitioning and node prediction framework. First, we employ a Weighted Neighborhood Rough Set algorithm to eliminate redundant features and screen for key diagnostic attributes. Second, based on the reduced attributes, we construct a patient similarity network and utilize the Louvain algorithm for initial community partitioning. Third, to enhance precision, Fuzzy Rough Set theory is introduced to handle boundary nodes. By defining the distance between nodes and communities, this step resolves the ambiguity of edge node classification. Finally, we incorporate the PageRank algorithm to predict the community attribution of new patients. The proposed method is applied to the field of Chronic Kidney Disease treatment. Experimental results on real clinical data demonstrate the scientific rigor and superior performance of the proposed framework compared to traditional methods.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 7","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147829330","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SFMamba: a novel spatial-frequency collaborative learning for multimodal medical image fusion with mamba","authors":"Zhaijuan Ding,&nbsp;Zhaisheng Ding,&nbsp;Yunzhe Men,&nbsp;Yanyu Liu,&nbsp;Shengyang Luan,&nbsp;Shufang Tian","doi":"10.1007/s10489-026-07252-8","DOIUrl":"10.1007/s10489-026-07252-8","url":null,"abstract":"<div><p>The objective of medical image fusion is to discern and amalgamate complementary features extracted from multimodal medical images, producing fused representations that are more interpretable and informative. This enhancement facilitates higher diagnostic accuracy and efficiency. While Transformer-based fusion methods excel at capturing long-range dependencies and enable parallel computation, their computational cost rises sharply with increasing input dimensions, often reaching quadratic complexity. To address this challenge, we propose a novel spatial-frequency domain fusion network, termed as SFMamba, which exploits a frequency transformation in conjunction with the Mamba model to fully leverage both spatial and frequency information. An efficient Mamba branch incorporates a spatial-frequency state-space (SFSS) model, reducing computational burden to linear or near-linear complexity. The selective-band feature extraction (SBFE) branch is constructed using a discrete wavelet pyramid, designed to capture multi-scale frequency components consistently across source images. To dynamically and effectively fuse multi-modal complementary information, we introduce a multi-domain feature fusion (MDFF) module that elevates fusion performance. Training is conducted with a multi-teacher learning strategy (MTLS) that integrates pre-trained convolutional neural networks and transformer-based fusion methods to generate multiple pseudo-labels, guiding the network to inherit fused knowledge from these priors. Extensive experiments demonstrate that SFMamba achieves state-of-the-art performance in both subjective and objective evaluations. The code for SFMamba is available at https://github.com/DZSYUNNAN/SFMamba.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 7","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147829331","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A legal text semantic generation framework based on multimodal feature fusion and dynamic weight optimization","authors":"Songling Qian,&nbsp;Yun Liu,&nbsp;Zheng Yang","doi":"10.1007/s10489-026-07167-4","DOIUrl":"10.1007/s10489-026-07167-4","url":null,"abstract":"<div><p>Due to the strict structured characteristics of legal provisions themselves, their semantics are highly dependent on explicit logical frameworks. Semantic analysis of them requires capturing implicit associations in the context. Therefore, the main difficulty of this study lies in the collaborative processing of multimodal features such as the structured provisions of legal provisions, the contextual relevance of precedents, and the implicit semantics of judicial interpretations. Traditional methods mainly focus on enhancing the logical coherence and thematic relevance of the generated text, while neglecting the integration of multimodal features of legal texts, resulting in poor semantic consistency between the semantic generation results and legal provisions or judicial documents. To this end, a legal text semantic generation framework based on multimodal feature fusion and dynamic weight optimization is proposed. The cross-modal noise in legal texts is filtered through regularization logic, and the Conditional Random Fields (CRF) domain adaptive word segmentation mechanism is selected to handle the ambiguity of professional term segmentation, achieving the preprocessing of legal texts. Introduce the attenuation factor of legal revision time and adopt the dynamic Term frequency-inverse Document Frequency (TF-IDF) Feature Engineering and Bidirectional Encoder Representations from Transformers for Legal EXpertise, BERT-LEX Obtain the weights of vocabulary in legal texts to extract the hierarchical semantic features of legal texts. By using the gated attention mechanism to dynamically embed the legal knowledge graph, a knowledge constraint generation model is constructed to enhance the legal consistency of the text and output the semantic generation results of the legal text. The big data set of real estate laws from 2020 to 2024 was selected for experimental verification. The results show that the ROUGE-L value, Legal BLEU value and Law Score value of the semantics of the legal texts generated by this method are all higher than 0.8, effectively enhancing the semantic coherence and legal logical accuracy.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 7","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-05-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147829133","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Auto-Correct OCR: a novel method for enhancing character recognition accuracy through error correction","authors":"Yanlin Ruan,&nbsp;Xiaoye Wang,&nbsp;Zhiwei Wen,&nbsp;Hongliang Gao","doi":"10.1007/s10489-026-07247-5","DOIUrl":"10.1007/s10489-026-07247-5","url":null,"abstract":"<div><p>Optical character recognition (OCR) systems deployed in industrial manufacturing environments face the dual challenges of high generalization requirements and extremely low tolerance for recognition errors. To address these issues, this paper proposes the Auto-Correct OCR framework, which comprises a training-free base recognition module and a domain-adaptive post-processing module termed structure-aware correction (SAC). Unlike conventional approaches that rely on extensive retraining, the framework introduces a hierarchical architecture that decouples structure learning from correction rule learning. For fixed-length strings, SAC employs a two-stage design: a structure learner that captures stable position-level character-type constraints, followed by a correction rule learner that applies position-independent classifiers with sample weighting for selective correction. For variable-length strings, SAC incorporates a dual-encoder Transformer architecture that fuses OCR features with product name semantics via cross-attention, enabling structure-aware character correction. Extensive experiments on brewery and automotive parts datasets demonstrate the effectiveness of the proposed approach. For fixed-length inventory codes, cross-temporal validation achieves an exact match rate exceeding 99%. For variable-length specification strings, SAC improves character accuracy from 97.23% to 99.63% and exact match rate from 84.14% to 96.58%. By leveraging structural priors, the framework delivers high accuracy and reliability under limited hardware resources, offering strong potential for rapid deployment in industrial manufacturing applications.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 7","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-05-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147829109","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Evolutionary multitasking by dynamic adaptive knowledge transfer for high-dimensional feature selection","authors":"Yujia Wei,&nbsp;Hongkun Lin,&nbsp;Binbin Chen,&nbsp;Juehan Lu,&nbsp;Shiguo Huang,&nbsp;Xiaolin Li","doi":"10.1007/s10489-026-07257-3","DOIUrl":"10.1007/s10489-026-07257-3","url":null,"abstract":"<div><p>Evolutionary multitasking (EMT) has shown strong performance in high-dimensional feature selection (FS) by building implicit parallelism bridges between different tasks. However, the knowledge transfer in existing EMT-based FS methods has the following problems: the knowledge for each task dynamically changes throughout the evolutionary process; the quality of knowledge across tasks may also vary significantly in the same iteration. If the dynamic characteristics of the task are ignored and a static knowledge transfer mechanism is adopted, this can result in the transfer of useless or negative knowledge, thereby reducing knowledge transfer efficiency and potentially misguiding or disrupting the search direction of the target task. To this end, we propose a novel EMT-based framework, namely MTABC. Specifically, three filter-based methods are used to generate low-dimensional search landscapes with multiple types of knowledge to enhance search diversity and reduce the search space. Secondly, an adaptive weight-based knowledge transfer mechanism adjusts the weight of each task based on real-time performance improvement, ensuring flexible and efficient knowledge transfer. In addition, we employ a modified gbest-guided Artificial Bee Colony (GABC) algorithm as the core optimizer, which is further enhanced by a Differential Evolution (DE) based elite bees secondary search strategy during the onlooker bee phase to improve exploitation and discover better solutions. Extensive experimental results confirm that our MTABC framework surpasses state-of-the-art FS methods on 15 datasets. Furthermore, the contributions of each component of the proposed MTABC are verified through ablation experiments.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 7","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-05-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147829108","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Efficient learning for active wrist compensation strategy of ledge-climbing robot","authors":"Andi Alfian Kartika Aji,&nbsp;Chi-Ying Lin","doi":"10.1007/s10489-026-07245-7","DOIUrl":"10.1007/s10489-026-07245-7","url":null,"abstract":"<div><p>Ledge-climbing robots face challenges in maintaining stability during continuous locomotion, particularly due to yaw deviation caused by gripper slipping during gripper exchange. This deviation accumulates over time and can cause the robot to fall. In this work, we propose a robust locomotion framework that integrates Central Pattern Generators (CPG) with Reinforcement Learning (RL) to enable stable multi-cycle locomotion by compensating yaw-induced lateral drift. We utilize CPG to generate a base rhythmic trajectory and train a residual RL policy to actively compensate for uncertainties. Specifically, we introduce an active wrist compensation mechanism where the RL agent learns a reference-free policy to regulate the wrist joint, correcting yaw deviation by minimizing lateral gripper displacement. To ensure efficient and smooth learning, we employ Early Stopping Policy Optimization (ESPO) combined with Generalized State-Dependent Exploration (gSDE). Training converges in 2 hours on an NVIDIA RTX 5070 Ti using 4096 parallel simulation environments. Simulation results demonstrate that our approach successfully achieves robust continuous locomotion, preventing falls where the open-loop CPG baseline fails after approximately 2000 steps. Furthermore, we found that training with wider friction randomization extends the operational friction range to lower coefficients while simultaneously improving locomotion speed by 21% at nominal friction levels. Robustness tests in simulation under sensor noise, actuator latency, mechanical backlash, and sensor drift show that the policy remains effective at moderate perturbation levels, with performance loss at higher levels. The code used in this study is available at: https://github.com/machiningman/RL-wrist-compensation.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 7","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-05-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-026-07245-7.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147829527","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}