Complex & Intelligent Systems最新文献

{"title":"A few-shot learning-based dual-input neural network for complex spectrogram recognition system with millimeter-wave radar","authors":"Kaiyu Chen, Shaoxi Wang, Wei Li, Yucheng Wang, Cunqian Feng, Yannian Zhou, Jian Cao, Binfeng Zong, Minming Gu","doi":"10.1007/s40747-025-01848-2","DOIUrl":"https://doi.org/10.1007/s40747-025-01848-2","url":null,"abstract":"<p>Graph data-driven machine learning methods for human activity recognition (HAR) have achieved success recently using sufficient data. In the realm of everyday life, we encounter a notable challenge: the scarcity of labeled radar samples. This limitation is compounded by the stark disparities in data distribution between simulated and measured activity domains. In this article, a generalized graph contrastive learning framework (DSMFT-Net) incorporated with Boulic-Thalmann simulation model for few-shot HAR is proposed. DSMFT-Net combines a clustering strategy with contrastive learning to develop a robust, domain-invariant feature representation. Particularly, the method divided into two phases: single radar range Doppler spectrogram prototypical contrast, enhancing the classification discriminative features by improving the compaction of prototypes and instances within a domain. Then, cross prototypical contrast of simulated and measured radar range Doppler spectrogram domain, focuses on discovering domain-invariant features through prototype-instance matching and proximity exploration. Moreover, mutual information maximization ensures the reliability of predictions, while pseudo-label information aids in self-supervised contrastive pre-training by comparing positive and negative sample pairs. The effectiveness of the model is empirically validated through testing conducted in both open and complex office environments. The experimental results indicate that the proposed method achieves an average accuracy of 93.3% under 5-shot setting and 96.5% under 10-shot setting across six human activity recognition tasks. These findings highlight the effectiveness of the proposed method in achieving high performance even with limited labeled data.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"128 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-04-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143862908","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":"Quantitative estimation method for complex part surface defects based on multimodal information fusion","authors":"Rui Wang, Wei Du, Qingchao Jiang","doi":"10.1007/s40747-025-01874-0","DOIUrl":"https://doi.org/10.1007/s40747-025-01874-0","url":null,"abstract":"<p>Surface quality is critical for the performance of high-end equipment, with defects potentially leading to severe operational failures. Current defect detection methods face challenges: 2D imaging lacks the ability to capture scratch depth, limiting quantitative damage assessment, while 3D point cloud methods are costly and time-consuming, hindering scalability in manufacturing. This study proposes a multimodal defect detection system (MDDS) that merges the benefits of 2D imaging and 3D point clouds for comprehensive defect analysis on complex parts. Utilizing a binocular vision system with high-precision industrial cameras, the system captures detailed 2D images and generates 3D point clouds through advanced reconstruction techniques. We enhance the Faster R-CNN network to improve defect localization and feature extraction, establishing a mapping between 2D images and 3D data to pinpoint defect-specific areas accurately. Additionally, we introduce a novel feature extraction approach using normal vector aggregation and the Fast Point Feature Histogram (FPFH) descriptor, combined with fuzzy C-means clustering, to detect and quantify scratch defects. This method assesses defect dimensions and depth, enabling precise damage classification. Tested on aero-engine impeller parts, our approach has proven effective in identifying and quantifying scratch defects on complex industrial components. The results demonstrate the system’s applicability and efficiency, making it a viable solution for practical implementation in industrial environments.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"216 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-04-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143862907","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":"M2FNet: multi-modality multi-level fusion network for segmentation of acute and sub-acute ischemic stroke","authors":"Shannan Chen, Xuanhe Zhao, Yang Duan, Ronghui Ju, Peizhuo Zang, Shouliang Qi","doi":"10.1007/s40747-025-01861-5","DOIUrl":"https://doi.org/10.1007/s40747-025-01861-5","url":null,"abstract":"<p>Ischemic stroke, a leading cause of death and disability, necessitates accurate detection and automatic segmentation of lesions. While diffusion weight imaging is crucial, its single modality limits the detection of subtle lesions and artifacts. To address this, we propose a multi-modality, multi-level fusion network (M²FNet) that aggregates salient features from different modalities across various levels. Our method uses a multi-modal independent encoder to extract modality-specific features from images of different modalities, thereby preserving key details and ensuring rich features. In order to suppress noise while ensuring the best preservation of modality-specific information, we effectively integrate features of different modalities through a cross-modal encoder fusion module. In addition, a cross-modal decoder fusion module and a multi-modality joint loss are designed to further improve the fusion of high-level and low-level features in the up-sampling stage, dynamically utilizing complementary information from multiple modalities to improve segmentation accuracy. To verify the effectiveness of our proposed method, M²FNet was validated on two public magnetic resonance imaging ischemic stroke lesion segmentation benchmark datasets. Whether single or multi-modality, M²FNet performed better than ten other baseline methods. This highlights the effectiveness of M²FNet in multi-modality segmentation of ischemic stroke lesions, making it a promising and powerful quantitative analysis tool for rapid and accurate diagnostic support. The codes of M²FNet are available at https://github.com/ShannanChen/MMFNet.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"41 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-04-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143862909","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":"Unsupervised feature selection based on generalized regression model with linear discriminant constraints","authors":"Xiangguang Dai, Mingyu Guan, Facheng Dai, Wei Zhang, Tingji Zhang, Hangjun Che, Xiangqin Dai","doi":"10.1007/s40747-025-01873-1","DOIUrl":"https://doi.org/10.1007/s40747-025-01873-1","url":null,"abstract":"<p>Unsupervised feature selection (UFS) methods play a crucial role in improving the efficiency of extracting relevant information and reducing computational complexity in the context of big data analysis. Despite notable advancements in the field of unsupervised feature selection for large-scale datasets, many UFS methods still remain redundant and irrelevant features during the feature selection process. To tackle these challenges, we present a novel unsupervised feature selection method that leverages the generalized regression model with linear discriminant constraints to learn discriminant and effective features from the data. Benefited from this, the relationships and patterns within the high-dimensional data are retained in the reduced-dimensional feature space. We reformulate our proposed method as a multi-variable optimization problem that incorporates equality constraints. To efficiently solve this problem, we develop an algorithm that updates each variable alternately. Extensive experiments on six datasets among nine state-of-the-art methods on the clustering task are conducted to demonstrate the effectiveness of the proposed method.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"15 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143857268","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":"Adaptive integrated weight unsupervised multi-source domain adaptation without source data","authors":"Zhirui Wang, Liu Yang, Yahong Han","doi":"10.1007/s40747-025-01871-3","DOIUrl":"https://doi.org/10.1007/s40747-025-01871-3","url":null,"abstract":"<p>Unsupervised multi-source domain adaptation methods transfer knowledge learned from multiple labeled source domains to an unlabeled target domain. Existing methods assume that all source domain data can be accessed directly. However, such an assumption is unrealistic and causes data privacy concerns, especially when the source domain labels include personal information. In such a setting, it is prohibited to minimize domain gaps by pairwise calculation of the data from the source and target domains. Therefore, this work addresses the source-free unsupervised multi-source domain adaptation problem, where only the source models are available during the adaptation. We propose trust center sample-based source-free domain adaptation (TSDA) method to solve this problem. The key idea is to leverage the pre-trained models from the source domain and progressively train the target model in a self-learning manner. Because target samples with low entropy measured from the pre-trained source model achieve high accuracy, the trust center samples are selected first using the entropy function. Then pseudo labels are assigned for target samples based on a self-supervised pseudo-labeling strategy. For multiple source domains, corresponding target models are learned based on the assigned pseudo labels. Finally, multiple target models are integrated to predict the label for unlabeled target data. Extensive experiment results on some benchmark datasets and generated adversarial samples demonstrate that our approach outperforms existing UMDA methods, even though some methods can always access source data.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"138 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143857269","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":"DFL topology optimization based on peer weighting mechanism and graph neural network in digital twin platform","authors":"Nguyen Anh Tuan, Atif Rizwan, Sa Jim Soe Moe, Anam Nawaz Khan, Do Hyeun Kim","doi":"10.1007/s40747-025-01887-9","DOIUrl":"https://doi.org/10.1007/s40747-025-01887-9","url":null,"abstract":"<p>Decentralized federated learning (DFL) represents a distributed learning framework where participating nodes independently train local models and exchange model updates with proximate peers, circumventing the reliance on a centralized orchestrator. This paradigm effectively mitigates server-induced bottlenecks and eliminates single points of failure, which are inherent limitations of centralized federated learning architectures. However, DFL encounters significant challenges in attaining global model convergence due to inherent statistical heterogeneity across nodes and the dynamic nature of network topologies. For the first time, in this paper, we present a topology optimization framework for DFL that integrates a peer weighting mechanism with graph neural networks (GNNs) within a digital twin platform. The proposed approach leverages local model performance metrics and training latency as input factors to dynamically construct an optimized topology that balances computational efficiency and model performance. Specifically, we employ Particle Swarm Optimization to derive node-specific peer weight matrices and utilize a GNN to refine the underlying mesh topology based on these weights. Comprehensive experimental analyses conducted on benchmark datasets demonstrate the superiority of the proposed framework in achieving accelerated convergence and enhanced accuracy across diverse nodes. Additionally, comparative evaluations under IID and Non-IID data distributions substantiate the robustness and adaptability of the approach in heterogeneous learning environments, underscoring its potential to advance decentralized learning paradigms.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"43 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143857271","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":"Multi-objective recommendation system utilizing a multi-population knowledge migration framework","authors":"Liang Chu, Ye Tian","doi":"10.1007/s40747-025-01891-z","DOIUrl":"https://doi.org/10.1007/s40747-025-01891-z","url":null,"abstract":"<p>Traditional recommendation systems tend to focus on accuracy and prefer recommending popular items, resulting in non-popular items rarely being exposed to users. However, recommending non-popular items to enhance users’ novelty experience is also crucial. Currently, many researchers are dedicated to multi-objective recommendation studies. Nevertheless, existing multi-objective recommendation algorithms often exhibit poor performance on the hypervolume value (HV) metric and lack effective methods to enhance novelty within evolutionary strategies. In this paper, we propose an innovative multi-objective recommendation algorithm based on a multi-population auxiliary evolution framework, abbreviated as MOEA-MIAE. Within this framework, we design three distinct optimization paths aimed at enhancing the convergence performance of the multi-objective algorithm and improving the hypervolume value metric of results. In addition to adopting the classical genetic algorithm as the main evolutionary population, we specifically introduce two auxiliary evolutionary populations. The first auxiliary population employs an HV-based multi-parent crossover method, while the second focuses on increasing the likelihood of generating highly novel solutions during crossover operations. These three evolutionary populations achieve effective complementarity and integration of their strengths through a mutual migration strategy of solution sets. Experimental results demonstrate that the proposed model exhibits superior performance in balancing accuracy and novelty, outperforming other comparable algorithms.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"108 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143857270","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":"Contrastive cross-domain sequential recommendation with attention-aware mechanism","authors":"Wei Zhao, Bo Li, Xian Mo","doi":"10.1007/s40747-025-01856-2","DOIUrl":"https://doi.org/10.1007/s40747-025-01856-2","url":null,"abstract":"<p>Cross-domain sequential recommendation (CDSR) aims to predict future sequential interactions in a target domain by analyzing historical sequence data from different domains. A significant challenge in CDSR is the accurate capture of user preferences based on the target domain and multiple domains. Existing methodologies to enhance the performance of the target domain primarily focus on learning preferences for a single domain within each respective domain and subsequently transferring this knowledge to the target domain via a transferring module. However, this approach inadequately accounts for the linear relationship between the target domain and user preferences, thereby limiting the potential benefits of leveraging target domain knowledge to enhance performance in rich domains. This study introduces a novel Contrastive cross-domain sequential recommendation technique with an attention-aware mechanism (<span>(hbox {C}^2hbox {DSRA}^2)</span>) for CDSR. We use graph neural networks (GNNs) combined with attention-aware mechanisms to elucidate the relationship between cross-domain and target domain user preferences. Specifically, we first develop an attention-aware framework over GNNs to capture collaborative relationships among inter-sequence items, then propose an attenuation function to assess the rationality of item representations. We construct cross-domain representations using the attention-aware mechanism to derive user-specific target domain representations. <span>(hbox {C}^2hbox {DSRA}^2)</span> enhances recommendation performance and practical applicability. Experiments show <span>(hbox {C}^2hbox {DSRA}^2)</span> surpasses state-of-the-art (SOTA) cross-domain recommendation algorithms.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"64 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143857211","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 adaptive dual distillation framework for efficient remaining useful life prediction","authors":"Xiang Cheng, Jun Kit Chaw, Shafrida Sahrani, Mei Choo Ang, Saraswathy Shamini Gunasekaran, Moamin A. Mahmoud, Halimah Badioze Zaman, Yanfeng Zhao, Fuchen Ren","doi":"10.1007/s40747-025-01886-w","DOIUrl":"https://doi.org/10.1007/s40747-025-01886-w","url":null,"abstract":"<p>Predicting the Remaining Useful Life (RUL) of industrial equipment is essential for proactive maintenance and health assessment, particularly under the computational constraints of edge devices. While deep learning methods, such as Long Short-Term Memory (LSTM) networks, excel at modeling complex time series, their high computational cost often restricts real-time deployment. To address this challenge, we present an Adaptive Dual Distillation Framework (A-DDF) that transfers knowledge from a large LSTM teacher model to a lightweight bidirectional Gated Recurrent Unit (GRU) student model. Soft-target distillation refines predictive distributions to provide robust supervision and our correlation-based feature alignment preserves inter-feature relationships and prevents information loss. An adaptive weighting mechanism balances these two distillation strategies, enabling the student model to maintain high predictive accuracy while reducing model complexity. We validate our approach on NASA’s C-MAPSS dataset, which includes diverse operating conditions. A-DDF outperforms previous methods, achieving a 12% decrease in relative error (MAPE), improving prediction accuracy and stability. Ablation experiments show the dual distillation strategy improves predictive accuracy, surpassing single distillation approaches. Notably, the student model achieves a 5.34-fold compression rate, reducing parameters by 83%, while maintaining or exceeding the performance of the LSTM teacher model. These results highlight A-DDF’s potential for efficient, high-accuracy predictive maintenance on edge devices. Comparisons with mainstream benchmarks confirm A-DDF’s superior performance across datasets. Finally, generality and quantization experiments validate its broad applicability and deployability. The proposed method emphasizes reducing model size without sacrificing performance, making it ideal for real-world predictive maintenance scenarios and intelligence-driven manufacturing applications.</p>","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"1 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143857267","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":"Encoding local label correlations in multi-instance multi-label learning with an improved multi-objective particle swarm optimization","authors":"Xiang Bao, Fei Han, Qinghua Ling","doi":"10.1007/s40747-025-01854-4","DOIUrl":"https://doi.org/10.1007/s40747-025-01854-4","url":null,"abstract":"<p>Label correlations, as important prior information, are essential to enhance the classification performance in Multi-Instance Multi-Label (MIML) algorithms, but existing models always leverage global label correlations which are less informative. Furthermore, classifier optimization is also crucial for MIML classification results, previous works do not frequently seek to optimize multi objectives simultaneously which may easily result in poor performance on some important metrics. In this paper, a MIML algorithm encoded with local label correlations with an improved Multi-Objective Particle Swarm Optimization (MIML-MOPSO-LLC) is proposed to address the above problems. Specifically, a framework is proposed by taking consideration into both global discrimination fitting and local label correlation sensitivity in the bag level simultaneously in the standard MIML. Subsequently, the loss function of the framework is solved by an alternating optimization process where Support Vector Machine (SVM) classifiers are constructed for optimization. Ultimately, an improved MOPSO is employed to optimize the SVM classifiers by searching for more reliable Pareto front solutions. The experimental results demonstrate that the proposed method achieves competitive performance compared with the classical and state-of-the-art MIML models from the perspective of several classification indicators. Notably, the proposed method which explores the local label correlations exhibits superiority over methods relying on global ones. Furthermore, the study reveals that proposed methods demonstrates enhanced effectiveness in MIML classification and optimizing SVM classifiers compared to conventional single and multi objective optimization approaches.</p><h3 data-test=\"abstract-sub-heading\">Graphical Abstract</h3>\u0000","PeriodicalId":10524,"journal":{"name":"Complex & Intelligent Systems","volume":"47 1","pages":""},"PeriodicalIF":5.8,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143857273","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}