Forthcoming Articles
International Journal of Information and Communication Technology
Forthcoming articles have been peer-reviewed and accepted for publication but are pending final changes, are not yet published and may not appear here in their final order of publication until they are assigned to issues. Therefore, the content conforms to our standards but the presentation (e.g. typesetting and proof-reading) is not necessarily up to the Inderscience standard. Additionally, titles, authors, abstracts and keywords may change before publication. Articles will not be published until the final proofs are validated by their authors.
Forthcoming articles must be purchased for the purposes of research, teaching and private study only. These articles can be cited using the expression "in press". For example: Smith, J. (in press). Article Title. Journal Title.
Articles marked with this shopping trolley icon are available for purchase - click on the icon to send an email request to purchase.
Online First articles are also listed here. Online First articles are fully citeable, complete with a DOI. They can be cited, read, and downloaded. Online First articles are published as Open Access (OA) articles to make the latest research available as early as possible.
Articles marked with this Open Access icon are Online First articles. They are freely available and openly accessible to all without any restriction except the ones stated in their respective CC licenses.Register for our alerting service, which notifies you by email when new issues are published online.
International Journal of Information and Communication Technology (34 papers in press) Regular Issues
Abstract: Addressing the issue that current cultural and creative product generation methods fail to account for user emotional needs, resulting in poor image generation outcomes, this paper first employs natural language processing algorithms to automatically segment user profiles and extract demand characteristics. Deep learning algorithms are introduced to analyse the sentiment behind user demands, thereby identifying emotional inclinations expressed by users. Building upon this foundation, a novel residual block architecture is designed with a diffusion model as the core network. The noise estimation network is enhanced by incorporating a convolutional block attention module. By integrating conditional control and user profiles as the control network, the approach effectively generates cultural and creative product images that align with users emotional expectations. Experimental results demonstrate that the proposed method achieves at least an 8.63% improvement in peak signal-to-noise ratio, enabling the generation of high-quality cultural and creative product images. Keywords: cultural and creative product generation; user profiling; conditional diffusion model; attention mechanism; natural language processing. DOI: 10.1504/IJICT.2025.10075148
Abstract: To address the challenges of high dimensionality, nonlinearity, and small sample size in traditional Chinese medicine data, this study proposes a novel data processing strategy using a particle swarm optimised deep belief network. The method automates deep belief network hyperparameter tuning via particle swarm optimised to enable end-to-end feature learning and classification. On the traditional Chinese medicine systems pharmacology Danshen blood-activation dataset, particle swarm optimised deep belief network achieved an accuracy of 87.8% and an F1-score of 86.8%, surpassing both conventional models, (e.g., XGBoost at 85.2%) and unoptimised deep belief network (83.1%). It also attained 98.8% accuracy on the University of California Irvine Wine dataset, demonstrating strong generalisation. This work offers an automated, high-precision computational tool for traditional Chinese medicine data analysis, significantly enhancing model performance and interpretability. Keywords: particle swarm optimisation; deep belief networks; DBNs; traditional Chinese medicine data; hyperparameter optimisation; feature learning. Abstract: Existing English speaking teaching path methods ignore learners behavioural characteristics, leading to inaccurate generation of personalised teaching paths. This paper first integrates the production-oriented approach to design the teaching process and constructs an intelligent teaching path generation model that integrates learner behavioural preference analysis and the teaching process. The model utilises graph convolutional networks and long short-term memory network to capture semantic associations and learners dynamic evolutionary characteristics. Then, using non-dominated sorting genetic algorithm II for multidimensional optimisation, multiple paths are optimised for multiple objectives to obtain a set of frontier solutions, and the English speaking teaching path with the highest score is obtained. Experimental results show that the suggested approach improves prediction accuracy by an average of 7.0423.96% while significantly reducing the time required to solve for the optimal path, validating the models efficiency. Keywords: English speaking teaching path; production-oriented approach; graph convolutional network; long short-term memory network; NSGA-II method. DOI: 10.1504/IJICT.2025.10075149
Abstract: This research proposes a CLST architecture that integrates multiple data sources using Siamese Neural Networks (SNN) to identify unusual financial transactions. By leveraging spatial, temporal, and multimodal feature learning alongside class imbalance handling, the model outperforms existing methods in recall, F1-score, and precision, enabling a robust early-warning system for fraud prevention. Multisource data fusion enhances detection accuracy by combining complementary information from diverse financial streams. While prior studies have applied rule-based, or machine learning methods to unimodal datasets, and recent multimodal approaches show promise, challenges remain in complex financial networks. The proposed hybrid method combines CNNs, LSTMs, MLPs, and SMOTE to address class imbalance, with SNN-based feature extraction improving robustness. Experiments demonstrate maximum precision of 0.937 and an F1-score of 0.787, with SNN + RF and SNN + SVM outperforming traditional and SMOTE-based models. Statistical analysis confirms SNN-based models achieve superior stability and balanced accuracy in anomaly detection. Keywords: multisource financial data fusion; anomaly transaction detection; early-warning mechanism; multimodal learning; fraud detection; credit card transactions. DOI: 10.1504/IJICT.2025.10075205
Abstract: This study builds a high-quality Miao pattern dataset, then applies label smoothing and mosaic data augmentation. To maximise multi-scale feature fusion, the spatial pyramid pooling fast (SPPF) module is utilised. Increasing the precision of bounding box regression and small target recognition, the focal loss and complete IoU Loss algorithms are combined. A web-based visual teaching platform is created with features for displaying cultural knowledge and inferring models. The research results indicate that, the enhanced YOLOv5 model outperforms comparable models like faster R-convolutional neural network and YOLOv4 with mean average precision@0.5 of 89.6% and mAP@0.5:0.95 of 61.5% on the test set. Compared to the original YOLOv5s, it has increased by 5.4% and 7.2% respectively. Meantime, the recall rate improvement in small pattern detection is greater than 6%, which is better than that of baseline models such as YOLOv4. The data confirm deep learnings potential in high-precision ethnic culture recognition and instruction. Keywords: Miao costume patterns; YOLOv5; target detection; transformer attention mechanism; intelligent teaching system. DOI: 10.1504/IJICT.2025.10075238
Abstract: This paper explores the application of blockchain technology in human resource management, focusing on employee performance appraisal and compensation management. The study presents a comprehensive design and implementation of a blockchain-based system that addresses the challenges of traditional HR processes. The proposed system leverages smart contracts, cryptographic techniques, and decentralised architecture to ensure transparency, security, and efficiency in managing employee performance and compensation. The research includes a detailed analysis of the systems architecture, smart contract design, and data management strategies. Performance evaluation and security assessments demonstrate the systems capabilities in handling high transaction volumes, ensuring data integrity, and resisting common attacks. While acknowledging limitations such as scalability constraints and regulatory challenges, the paper highlights the potential of blockchain technology to revolutionise HR management practices. The study concludes by outlining future research directions, including system optimisation, integration with other HR modules, and the exploration of advanced analytics techniques. Keywords: blockchain; human resource management; performance appraisal; compensation management; smart contracts; decentralised systems; data security; HR analytics. DOI: 10.1504/IJICT.2025.10075239
Abstract: How to achieve objective teaching process identification and real-time teacher ability evaluation has become an important direction of educational informatisation research. This paper puts forward a model of teaching behaviour recognition and ability dynamic evaluation, which combines 3D-CNN and Bi-LSTM, and realises automatic recognition of classroom teaching behaviour and continuous quantitative analysis of teachers ability through visual data. The experimental results show that the proposed model achieves 93.6% accuracy, 91.7% recall and 0.92 F1-value in the task of teaching behaviour recognition, which is significantly better than the traditional convolution or single time series model. The dynamic evaluation module realises the real-time tracking of teachers ability through the time-weighted mechanism, and the coincidence rate between the system and the manual expert score reaches 91.8%. The research results finds this method boosts the intelligent level of teaching process monitoring, promote the transformation from subjective experience evaluation to data-driven dynamic evaluation. Keywords: image sequence mining; teaching behaviour recognition; dynamic capability evaluation. DOI: 10.1504/IJICT.2025.10075240
Abstract: Pragmatic ability assessment holds significant importance in language teaching and related fields, yet existing methods fail to capture and utilise the characteristics and information across different modalities. To address this, this paper optimises graph neural networks through multi-stage adaptive fusion. By decomposing the graph neural network into a multi-stage training format, higher-order features of graph data are progressively integrated into shallow models across multiple stages, thereby training a more robust shallow model. Subsequently, a pragmatic competence prediction model based on an improved graph neural network and multi-feature fusion is proposed. First, modal information is progressively integrated to ensure comprehensive fusion. Then, long-range pragmatic information is captured and incorporated into sentence-level information, enabling the model to better understand global features. Experimental results demonstrate that the proposed model achieves at least a 3.46% improvement in pragmatic competence test accuracy, facilitating more precise assessment of pragmatic competence levels. Keywords: pragmatic competency assessment; graph neural network; GNN; multimodal feature; multi-stage optimisation; adaptive fusion. DOI: 10.1504/IJICT.2025.10075259
Abstract: This paper addresses the task of emotion recognition in childrens drama performances by proposing an attention-based multimodal feature fusion model. The model extracts fine-grained facial expression features from the visual modality using a pre-trained deep network, and derives Mel-spectrograms and acoustic parameters from the audio modality. These feature streams are then dynamically calibrated and integrated via a cross-modal attention fusion module to capture key emotional cues in dramatic contexts. Evaluated on the public RAVDESS dataset of dramatised speech clips, our model achieves a weighted accuracy of 79.4% and an F1-score of 0.782, demonstrating a significant improvement over feature concatenation-based baseline fusion methods. The results indicate that the model effectively perceives subtle emotional dynamics in theatrical settings, offering a reliable tool for childrens affective computing. Keywords: multimodality; children's theater; emotion recognition; attentional mechanisms. DOI: 10.1504/IJICT.2025.10075260
Abstract: With the increasing prominence of psychological health problems of students in colleges and universities, efficient and accurate identification of psychological states has become an urgent issue. To address this issue, this paper proposes a psychological monitoring system for college students based on affective computing and behavioural trajectory analysis (AffectPath-PM). The system firstly extracts students multimodal characteristics by using the affective computing module, secondly obtains the characteristics of different patterns through the analysis of behavioural trajectories, and finally real-time monitoring and intervention is achieved by combining comprehensive assessment and early warning feedback. Experimental results indicate that the overall performance of this system demonstrates an average improvement of approximately 4.5% compared to the reference method. Small-scale validation experiments also demonstrate its applicability and scalability. This system offers universities a comprehensive and efficient mental health monitoring solution, possessing significant practical value. Keywords: affective computing; behavioural trajectory analysis; psychological health monitoring; multimodal data. DOI: 10.1504/IJICT.2025.10075261
Abstract: Personalised sports training has emerged as a critical component in optimising athletic performance and minimising injury risks. Nevertheless, conventional approaches predominantly depend on coaches subjective expertise, which often falls short in delivering dynamically precise adaptations. In response, this study introduces a reinforcement learning-based framework for generating individualised training regimens. By formulating the training process as a Markov decision process, the system enables an intelligent agent to interact with a simulated training environment, producing optimised training actions derived from real-time user status information. Evaluations conducted on the public FitRec dataset indicate that, relative to conventional baseline techniques, the proposed system yields an average improvement of 15% in predicted performance indicators, while concurrently lowering the incidence of training overload by 30%. These findings highlight the potential of the proposed framework as an effective new paradigm for automated and individualised sports science training. Keywords: reinforcement learning; RL; personalised sports training; Markov decision process; MDP; reward function; FitRec dataset. DOI: 10.1504/IJICT.2025.10075262
Abstract: University libraries are confronted with the challenges of low resource utilisation rate and insufficient modelling of the dynamic evolution of readers cognition. Traditional collaborative filtering methods are difficult to quantify cognitive state changes and ignore the influence of environmental factors on resource adaptability. To this end, this study proposes a dynamic recommendation model that integrates multi-factor knowledge tracing (MFKT) and graph neural networks (GNN). The reader cognitive state matrix is constructed through gated recurrent unit (GRU) time series modelling. Combined with behavioural pattern analysis and environmental feedback mechanism, the dynamic balance of resource difficulty and popularity is achieved. The cognitive graph convolutional network (CGCN) is designed based on the Pareto optimality theory to synchronously optimise the recommendation accuracy, knowledge gain and resource coverage. This study provides a referable technical solution to solve the problem of accurate matching between resources and readers cognition. Keywords: multifactor knowledge tracking; MFKT; book recommendation model; graph neural networks; GNN; gated recurrent unit; GRU. DOI: 10.1504/IJICT.2025.10075263
Abstract: In this paper, an innovative early warning model integrating news sentiment analysis and deep learning is proposed to address the complexities of regional economic risk identification and early warning. The model extracts spatio-temporal features from macroeconomic indicators and news texts respectively through a dual-channel network structure, and utilises the attention mechanism for dynamic fusion. Experiments based on Chinas provincial panel data and global database of events, language and tone news data show that this model achieves the harmonic mean of precision and recall of 0.812, which represents a significant improvement of 8.9% over the best-performing benchmark model (XGBoost at 0.745) and 16.3% over the traditional logistic regression model (0.698). Furthermore, this model can identify potential risk areas earlier. These findings provide new methods and decision support technologies for regional economic risk monitoring, which are of great significance to policymakers and financial regulatory authorities. This study is validated on Chinese provincial data, and generalisability to other regions requires further testing. Future work will explore finer spatial granularities and diverse data sources. Keywords: regional economic risk early warning; sentiment analysis; deep learning; attention mechanisms; multi-source data fusion. DOI: 10.1504/IJICT.2025.10075264
Abstract: Emotional intervention plays a crucial role in mental health support, yet traditional approaches often lack the dynamic adaptability to individual states and contextual changes. To address these limitations, this study proposes a value-guided meta-adaptive reinforcement learning framework. By integrating meta-learning with deep reinforcement learning, this approach enables intervention strategies to rapidly adapt to users real-time emotional states and long-term needs. We design an attention-based meta-policy network to extract shared representations across users and introduce a value function to quantify long-term psychological benefits. Furthermore, the framework employs proximal policy optimisation for policy training and dynamically adjusts hyperparameters through a meta-adaptive mechanism to handle non-stationary intervention environments. Experiments on simulated and real-world user datasets demonstrate that the proposed method achieves approximately 22% higher emotional improvement rates and 33% faster convergence speed compared to the best baseline. Keywords: meta-adaptive reinforcement learning; affective computing; personalised intervention; proximal policy optimisation; PPO. DOI: 10.1504/IJICT.2025.10075265
Abstract: Existing methods for tracking the evolution of ideological and political public opinion struggle to fully uncover intermodal correlations and exhibit low tracking accuracy. To address this, this paper first employs the Shapley additive explanations algorithm optimised by random forests to screen key influencing indicators. These selected indicators undergo a Gramian angular field transformation to generate a two-dimensional image. Subsequently, the Shapley additive explanations optimises the self-attention mechanism of the Transformer model while enhancing locally significant features that substantially influence tracking outcomes. The improved transformer model and bidirectional encoder representations from transformers model are employed to extract image and text features, respectively. Contrastive learning is introduced to align features across modalities. Multimodal fusion features undergo classification via the softmax function, enabling the tracking of public opinion evolution. Experimental results demonstrate that the proposed model achieves a tracking accuracy of 92.1%, exhibiting outstanding tracking efficiency. Keywords: ideological and political public opinion tracking; transformer model; SHAP algorithm; multimodal feature fusion; contrastive learning. DOI: 10.1504/IJICT.2025.10075266
Abstract: Basketball player action optimisation is key to improving competitive performance. To address the issue of insufficient mining of biomechanical features and poor action optimisation effects in current research, this paper first conducts biomechanical data analysis of basketball players and constructs a motion state equation for capturing and analysing key data. Then, it integrates image and biomechanical motion data to provide comprehensive multimodal perception information. A multimodal feature extraction and fusion module based on self-attention mechanism is designed. Secondly, the pose action decision-making task of the athlete is modelled as a deep reinforcement learning (DRL) problem. Finally, a hybrid reward function is designed to achieve efficient training of the model and action strategy optimisation. Experimental outcome indicates that the high model improves the action optimisation success rate by at least 5% compared to the baseline model, demonstrating good action optimisation effects. Keywords: basketball action optimisation; deep reinforcement learning; biomechanical modelling; multimodal feature fusion; attention mechanism. DOI: 10.1504/IJICT.2025.10075267
Abstract: Global climate change has made accurate urban carbon sink accounting crucial for low-carbon policies and ecological restoration, yet traditional methods suffer from inefficiency, low precision, and poor generalisation. To address these issues, this study proposes an edge IoT-causal graph neural network framework. It integrates a multi-layer edge internet of things architecture reducing data transmission latency by over 40% compared to cloud-centric systems. Additionally, a causal graph neural network model is developed: it infers the causal structure of environmental variables via an improved PC algorithm and embeds this structure into graph attention network training to avoid spurious correlations. Experimental validation on real urban green space data shows the framework achieves 94.7% accounting accuracy, outperforming traditional graph neural networks, support vector machines, and remote sensing inversion by over 8.5%. This work provides a practical technical paradigm for high-precision urban carbon sink accounting, supporting evidence-based urban low-carbon management. Keywords: edge IoT; causal graph neural network; CGNN; urban carbon sink accounting; causal inference; multi-source data fusion. DOI: 10.1504/IJICT.2025.10075268
Abstract: Oral diseases pose a global health challenge characterised by highly subjective diagnosis and a lack of intelligent decision-making tools, often exacerbated by fragmented data silos. This study aims to construct a comprehensive big data analytics platform and intelligent decision support system for oral health, enabling data-driven precision diagnosis and treatment through a unified four-layer architecture. The platform integrates multi-source heterogeneous data and employs advanced deep learning models for accurate caries segmentation and periodontitis risk prediction. Experiments on public datasets demonstrate a dice coefficient of 92.5% for caries segmentation and an area under the receiver operating characteristic curve value of 0.94 for periodontitis risk prediction, with results showing statistical significance. The system significantly enhances the automation and interpretability of oral disease analysis, providing a reliable and efficient tool for clinical diagnostic assistance and facilitating personalised treatment planning. Keywords: oral health big data platform; intelligent decision support system; deep learning; periodontitis risk prediction; clinical assisted diagnosis. DOI: 10.1504/IJICT.2025.10075269
Abstract: As brand visual content becomes more important in marketing, it has gotten harder to make images that fit the brands identity. This research presents a generative diffusion approach utilising knowledge graph improvement for brand visual generation to tackle this issue. By putting the brand knowledge graph into the generation process, the system gives semantic direction for making images. The first step in the plan is to construct a knowledge graph that shows the brands main qualities. Then, a generative diffusion model based on the knowledge graph is made and tested to see how well it works for brand visual production. The model enhances the inception score (IS) by 21.3% and diminishes the Frechet inception distance (FID) by 19.5% in comparison to the conventional generative model. The model makes pictures that are consistent with the brand and seem beautiful, with good brand customisation and innovation. Keywords: knowledge graph; generative diffusion model; brand visual generation; brand consistency; visual quality. DOI: 10.1504/IJICT.2025.10075270
Abstract: Knowledge tracking (KT) is a core task in the domain of integrated Chinese language education. However, traditional KT methods struggle to fully uncover the complex knowledge relationships within Chinese language education. To address this, this article designs a knowledge heterogeneous graph in the domain of Chinese language education, designs a heterogeneous graph neural network (GNN) to learn interactive relations among nodes, and extracts exercise node features as exercise representations. Then, a deep residual network is suggested to learn the interaction among exercise representations and students answering abilities. Finally, a temporal convolutional network is used to track students cognitive states and forecast the probability of them correctly answering the next exercise. Experimental results on the ASSIST and KDD datasets show that the proposed method improves prediction accuracy by at least 2.74% and 3.41%, respectively, enabling accurate forecasting of the mastery level of Chinese language knowledge points. Keywords: Chinese language education integration field; knowledge tracking; graph neural network; GNN; deep residual network; temporal convolutional network; TCN. DOI: 10.1504/IJICT.2025.10075271
Abstract: Singing technique assessment is a crucial component in enhancing the quality of music education. To address the issue of insufficient assessment accuracy caused by the coupling of time-frequency features in existing methods, this paper first performs preprocessing on singing audio to extract time-frequency features. Then, by combining deep separable convolutions with dilated convolutions, it simultaneously models frequency and temporal features. Additionally, a residual network is employed to mitigate the gradient vanishing problem in deep network structures. Second, a spatio-temporal enhancement branch is constructed based on a bidirectional long short-term memory (BiLSTM) network. Through a gating mechanism, decoupled features are bidirectionally transmitted between temporal and frequency domains. Decoupled time-frequency feature sequences are then clustered to enable the model to intelligently evaluate singing segments. Experimental results show that the proposed model achieves at least a 4.71% improvement in evaluation accuracy, demonstrating a significant advantage over baseline models. Keywords: singing technique evaluation; time-frequency feature decoupling; deep separable convolution; bidirectional long short-term memory model; feature clustering. DOI: 10.1504/IJICT.2025.10075272
Abstract: With the development of society and the improvement of peoples living standards, swimming and fitness have gradually become important activities in daily life. To improve the safety management of swimming facilities, this study proposes a YOLOv7 and DeepSORT algorithm for abnormal swimming behaviour detection and multi-object tracking. This method first uses YOLOv7 for object detection, and then continuously tracks the detected targets through the DeepSORT algorithm. To optimise feature extraction for small targets, this study utilises spatial pyramid deformable convolution module and non-local attention module attention mechanism for improvement. In addition, to further improve tracking accuracy, DeepSORT introduces distance intersection and union ratio. The results showed that the improved object detection accuracy, recall and F1-value reached 94.56%, 93.89%, and 95.08%, respectively. The accuracy of multi-object tracking on the training and testing sets reached 88.56 and 90.54, with an improved accuracy value of 89.42. In addition, the detection rate of the research method exceeded 86% in crowded scenes and 91% in sparse scenes, with a minimum false alarm rate of only 1.2 times per hour. The constructed method can identify and track abnormal swimming behaviour, providing technical support for pool safety management. Keywords: YOLOv7; DeepSORT; behaviour detection; multi-object tracking; swim. DOI: 10.1504/IJICT.2025.10075292
Abstract: Currently, optimising teachers teaching paths using graph convolutional networks has become an important exploration direction for improving teaching quality. However, there are still issues with poor model performance and an imperfect evaluation system. To this end, this paper optimises the model algorithm to enhance the research findings of GCN-based optimisation of teaching paths for university teachers. Firstly, this paper explores the potential connections between data by redesigning convolution kernels to adapt to the complex structure of teaching data. At the same time, this paper uses efficient parameter update algorithms such as adaptive moment estimation to dynamically adjust the learning rate based on the first-order and second-order moment estimates of the parameters, in order to accelerate model convergence. The research results indicate that the improved GCN model has an accuracy of 0.97, a precision of 0.96, and a training time of 12 hours when recommending teaching resources. Keywords: college education; teaching path optimisation; GCN model; teacher teaching; student development. DOI: 10.1504/IJICT.2025.10075293
Abstract: This study addresses real-time vehicle management challenges in instant delivery logistics, where conventional methods face dynamic demands causing inefficiencies and higher costs. It proposes the elastic dynamic scheduling algorithm (EDSA), integrating real-time data (e.g., delivery requests, traffic) and a multi-objective genetic algorithm optimising delivery time, operational costs, and energy consumption. EDSA dynamically adjusts routes amid disruptions like congestion or new orders. Simulations compare it with PPO-DRL, HGA, and MOPSO, showing superior performance: lower average delivery time, cost, and energy use; 12.2% higher peak delivery efficiency than PPO-DRL; and 5.6% reduced CO emissions versus HGA. This research fills gaps in existing methods via real-time adaptability and multi-objective optimisation, offering a scalable, holistic solution balancing cost, time, and environmental impact, providing actionable insights for logistics firms. Keywords: elastic dynamic scheduling algorithm; EDSA; instant delivery logistics; multi-objective optimisation; genetic algorithm; energy efficiency; real-time. DOI: 10.1504/IJICT.2025.10075294
Abstract: This study proposes a method for 3D facial modelling and animation generation integrated with emotion recognition. This method deeply combines high-precision emotion recognition with animation driving to improve the naturalness of facial expressions, emotional accuracy, and real-time interaction capabilities. Experimental results show that the average accuracy rate of 3D facial emotion recognition increases from 89.1% to 92.1%, with happiness (HA) reaching 98.9%, verifying the models high reliability and stability. In animation generation experiments, the emotion recognition accuracy reaches 90%, emotional consistency is 88%, and the average frame generation time is 48 ms/frame, all outperforming the control model. The research innovation and contribution lie in proposing a systematic integration strategy of emotion recognition with 3D modelling and animation generation. This enriches the theoretical framework of facial animation generation, achieving a balance among accuracy, naturalness, and efficiency, and providing an efficient and feasible technical solution. Keywords: emotion recognition; 3D facial image; facial modelling; animation generation. DOI: 10.1504/IJICT.2025.10075321
Abstract: With the rapid development of online education, improving students learning efficiency and experience has become a key research area. This study aims to address the challenges of predicting student behaviour and optimising learning paths on online education platforms. We propose a patented model that combines Dijkstras algorithm with the ant colony optimisation algorithm to predict student behaviour and optimise learning paths. The experimental results show that the model significantly improves the prediction accuracy, with an accuracy rate of 85.3%. In addition, after path optimisation, the learning efficiency increased by 20%, proving the effectiveness of the model in improving student performance. This study contributes to the development of personalised teaching methods by optimising students learning paths through the use of intelligent algorithms and presents a patented solution for the intelligent development of online education platforms. Keywords: Dijkstra algorithm; ant colony optimisation algorithm; student behaviour prediction; learning path optimisation. DOI: 10.1504/IJICT.2025.10075322
Abstract: In this study, we provide a system that shows students how to play piano with improvisation and accompaniment using cloud computing, deep learning, and CNN. Automatic evaluation of performance aspects, such as pitch, timbre, articulation, rhythm, and dynamics, is one way the suggested approach enhances piano lessons. Applying a hybrid approach that combines a matched filter with a rapid guided filter optimises preprocessing for music feature extraction. To further improve the accuracy of piano performance analysis, attention-induced multi-head CNNs and perceptual evaluation datasets are employed. In adaptive and remote learning settings, the technique shows better dependability and scalability. The model successfully integrates visual and aural methods of teaching piano, supports multilevel perceptual feature analysis, by providing a novel framework that enhances learning outcomes, enables tailored instruction, and adapts to the diverse needs of learners, this research contributes to the expanding field of intelligent music education. Keywords: artificial intelligence; AI; piano teaching; improvisation accompaniment; convolutional neural network; CNN; deep learning; DL; cloud computing; perceptual features; music education technology. DOI: 10.1504/IJICT.2025.10075323
Abstract: In this study, a multi-objective optimisation framework incorporating migration learning is proposed with the aim of efficiently optimising micro-milling parameters and accurately predicting surface roughness. First, a deep neural network (DNN)-based surface roughness prediction model is constructed as a base model. Subsequently, the pre-trained model is fine-tuned (fine-tuning) using a limited amount of micro-milling experimental data in the target domain to quickly adapt to the target working conditions and significantly improve the prediction accuracy under small samples. On this basis, the migration learning-enhanced prediction model is integrated with a multi-objective optimisation algorithm (e.g., NSGA-II) to construct an optimisation framework. Experimental results show that relying on the millisecond evaluation capability of the migration learning agent model and the improved search strategy of NSGA-II, the Pareto frontier distribution range is widened by 28% and the frontier convergence speed is improved by 42%. Keywords: migration learning; micro-milling; multi-objective optimisation; surface roughness prediction. DOI: 10.1504/IJICT.2025.10075341
Abstract: This paper aims to address these issues by proposing a fine-grained classification algorithm system that integrates multi-level support vector machines (SVMs) and feature engineering. This system combines data pre-processing, feature enhancement, and multi-level SVM classification to construct a hierarchical decision-making structure of genre clusters genres periods. Experiments show that the algorithm achieves an accuracy of 92.3% on a diverse dataset, with a macro F1 score of 0.915. Furthermore, in robustness tests, the accuracy only drops to 88.0% under noise perturbation, 87.3% under blur perturbation, and 85.5% under occlusion perturbation. The cross-dataset generalisation accuracy reaches 85.2%. External validation indicates an average accuracy of 82.3% for non-Western oil paintings, while also demonstrating high interpretability. This paper contributes an innovative technical approach for fine-grained classification of oil painting genres; it enhances accuracy, robustness, and interpretability, and lays the foundation for subsequent lightweight design, multimodal fusion, and cross-media expansion research. Keywords: multi-level SVM; support vector machines; feature engineering; oil painting; genre; classification.
Abstract: Ice accumulation on insulators can lead to electrical breakdown, equipment damage, and line outages, making timely and accurate detection essential for maintaining the safe and stable operation of power systems. This paper proposes an ice accretion detection method for insulators based on You Only Look Once version 11 (YOLOv11), integrating image processing and deep learning techniques to achieve automated detection. A self-built dataset was used to fine-tune YOLOv11, enhancing the model's accuracy and robustness in complex environments. Compared to its predecessors, YOLOv11 features an improved backbone network for more efficient feature extraction, advanced attention mechanisms for enhanced focus on critical regions, and an anchor-free detection head that reduces computational complexity while maintaining high precision. Multi-scale feature fusion ensures the accurate detection of ice accretion of various sizes, while dynamic label assignment optimises alignment between predictions and ground truth. Experimental results demonstrate that the fine-tuned YOLOv11-based algorithm achieves high mean average precision (mAP) and F1-scores on the test set, indicating robust detection performance. The proposed method not only enhances detection efficiency but also reduces labour costs, making it well-suited for large-scale power line monitoring. Keywords: ice accumulation; insulator icing detection; YOLOv11; ice accretion detection. DOI: 10.1504/IJICT.2025.10075001
Abstract: To address the usability challenges faced by elderly users when operating smart home systems in the context of an aging population, this study proposes a smart aging-friendly home control algorithm framework based on multimodal interaction. The core of this framework lies in the innovative design of three key algorithms: a multimodal fusion decision-making algorithm that integrates speech recognition, simple gesture understanding, and touches intent analysis; an aging-friendly interaction optimisation algorithm; and a context-aware intelligent assistance algorithm. The proposed algorithms were validated through user simulation and comparative experiments. The results indicate that the algorithm framework effectively improves elderly users' operational efficiency and task completion rates while significantly reducing cognitive load and operational error rates. This study provides core algorithmic support and practical design guidelines for constructing truly elderly-friendly smart home interaction systems. Keywords: multimodal interaction; age-friendly home; context awareness; smart home. DOI: 10.1504/IJICT.2025.10075002
Abstract: Artificial intelligence (AI) and internet of things (IoT) convergence brings immense opportunity to convert the laboratory environment into intelligent, adaptive systems. This study proposes an integrated AI-IoT framework for smart laboratory engineering construction and engineering management team optimisation, which overcomes the current shortcomings in resource efficiency, task scheduling, and environmental control to some extent. In this system, real-time IoT sensor networks monitor ecological and operational conditions; meanwhile, LSTM models are applied for predictive environmental control, genetic algorithms for dynamic task scheduling, and SVM classifiers for human activity recognition. The framework was deployed in a research laboratory for six months, and the system achieved substantial improvements: energy consumption was reduced by 28.48%, equipment downtime by 54.37%, and task overlap and average task duration were significantly minimised. Additionally, predictive maintenance accuracy reached approximately 93.2%, eliminating passive interventions and improving equipment availability. Since intelligent task allocation incorporates fault tolerance considerations, workload imbalance in task execution is alleviated, and staff satisfaction is enhanced. Our results demonstrate that a collaborative AI-IoT approach can effectively improve infrastructure efficiency and worker productivity. In this context, the proposed framework provides a scalable, sustainable, and context-aware solution for next-generation laboratory environments in academic and industrial domains. Keywords: artificial intelligence; AI; internet of things; IoT; smart laboratory; engineering management; predictive maintenance; task scheduling; environmental monitoring. DOI: 10.1504/IJICT.2025.10075003
Abstract: This paper presents advancements in electromechanical systems aimed at enhancing the precision and reliability of innovative product design through embedded sensing and intelligent control. A self-sensing electromechanical system is developed by integrating embedded time-grating displacement sensors into traditional mechanical structures, enabling high-resolution position detection of motors, worm gears, and bearings. Methods for current reconstruction using a single resistance sensor and error suppression through optimised light-field structures are proposed to improve detection accuracy. Experimental validation demonstrates significant improvements in measurement precision and error reduction through adaptive weighted data fusion. The findings highlight the potential of combining electromechanical design, sensing integration, and intelligent algorithms to support next-generation innovative product service systems (PSS). Keywords: electromechanical systems; smart product service systems; embedded sensors; time-grating displacement; error correction; data fusion; MEMS actuators; condition monitoring. DOI: 10.1504/IJICT.2025.10075082
Abstract: Temperature compensation is crucial for improving sensor accuracy and stability in high-precision measurement. Although radial basis function (RBF) neural networks perform well in nonlinear modelling, they face slow convergence, long training time, and limited accuracy. To address these issues, this paper proposes an improved RBF algorithm (QOLS-RBF) by combining quantum controlled-NOT (C-NOT) gates with orthogonal least squares (OLS) theory. The method quantises input data and applies quantum superposition, entanglement, and interference to enhance feature extraction and centre aggregation. It further integrates OLS screening with the maximum error compression ratio, using C-NOT gate evolution to reduce hidden layer nodes and accelerate convergence. Experiments with 85 training and 170 testing sensor datasets show that QOLS-RBF outperforms RBF, OLS-RBF, K-means RBF, and FCM-RBF in convergence speed, training time, error accuracy, and network compactness. This approach enables efficient temperature compensation and offers a promising tool for modelling complex nonlinear systems. Keywords: neural network algorithm; orthogonal least squares; OLS; sensors. DOI: 10.1504/IJICT.2025.10075004 |
Open Access
Keep up-to-date
-
Our Blog -
Join us on Bluesky -
Follow us on X -
Visit us on Facebook - Our Newsletter (subscribe for free)
- RSS Feeds
- New issue alerts
- Inderscience is a member of publishing organisations including:
-
-
