Forthcoming and Online First 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 Articles will not be published until the final proofs are validated by their authors.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 published online here, before they appear in a journal issue. 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.
Register for our alerting service, which notifies you by email when new issues are published online.
International Journal of Information and Communication Technology (20 papers in press) Regular Issues
Abstract: Cloud occlusion challenges remote sensing image processing by degrading quality and analysis accuracy. Existing cloud removal methods capture local features but struggle with global dependencies and cloud morphology, limiting detail restoration and consistency. To address this, we propose a multi-scale adaptive graph convolution generative adversarial network (MAGC-GAN), integrating a multi-scale adaptive graph convolution network (MAGCN) and an adaptive patch discriminator (APD). MAGCN enhances spatial dependencies using adaptive graph convolution, effectively reconstructing cloud-covered regions by capturing global contextual relationships. A multi-scale feature fusion mechanism enables adaptation to varying cloud thicknesses. APD improves fine-detail recovery by evaluating multiple local patches individually, using an adaptive affine transformation matrix. It also incorporates texture-aware and global consistency losses to restore high-frequency details while maintaining coherence. Compared to existing methods, MAGC-GAN significantly enhances cloud-occluded region restoration, particularly in detail recovery and precise cloud edge reconstruction. Keywords: remote sensing image; cloud removal; graph convolutional network; GCN; generative adversarial network; GAN; adaptive. DOI: 10.1504/IJICT.2025.10071097
Abstract: The inaccurate topology of low-voltage distribution station area leads to inefficient fault disposal and affects user experience. In view of the above situation, this paper proposes a data-driven full hierarchical topology identification method for low-voltage distribution station areas. The density peak K-means (DPK-means) algorithm is employed in this method to discern the users phase based on the similarity analysis of distribution transformer-branch box-metre box-user. Furthermore, the Kendall correlation coefficients between the voltage curves at each hierarchy in the low-voltage distribution station area are calculated and normalised to identify subordinate relations. The proposed method enables the recognition of user phases and hierarchical subordinate relations within the low-voltage distribution station area. Finally, the effectiveness of the proposed method is analysed and verified in the actual distribution station area. Keywords: low-voltage distribution station area; full hierarchy topology identification; the similarity of voltage curves; density peak K-means; DPK-means algorithm; Kendall correlation coefficient. DOI: 10.1504/IJICT.2025.10071098
Abstract: Ceramic art, deeply rooted in cultural heritage, has long been regarded as a symbol of craftsmanship and historical significance, often commanding substantial prices in the art market. However, with the rise of artificial intelligence (AI) and its ability to generate art that closely resembles human creations, distinguishing between authentic and AI-generated artworks has become a critical challenge. In this research work deep learning base models including the proposed convolutional neural networks (CNNs) and pre-trained models are applied to identify ceramic arts, distinguishing between human prepared artefacts and AI-generated content (AIGC). There is no benchmark data set available for the distinguish between real ceramic and AI-generated, therefore, the dataset has been prepared having two classes: authentic ceramic items (real) and AI-generated. The results obtained the highest accuracy by using CNN is 98%, by comparison of pre-trained models including ResNet, VGG and AlexNet model. This study may help to identify the authenticity of digital artefacts in the digital era. Keywords: deep learning; cultural heritage; ceramic; artificial intelligence; classification; computer vision; feature extraction; norm analysis.
Abstract: Neural machine translation (NMT) is the study endeavoured to build systems that would assimilate human language deciphering and production by utilising far-reaching linguistic and contextual forms. This article provides the details about an adaptive neural machine translation (ANMT) model which has incorporated the attentional mechanisms to deal with English texts. A proposed model is compared to existing best practice translation frameworks which are then included with two different approaches such as idiomatic expressions, domain-specific terminologies, and low-resource scenarios. We propose a new adaptation where user feedback loops are used as a method for refining translations based on emerging linguistic patterns. The experimental results confirm that ANMT was a success and the translation mistakes had lessened when new models were adopted; additionally, indicating that NMT experts had received a much better score compared to the baseline language model. This means that ANMT is a significant step in the evolution of AI technologies in translation. Keywords: neural machine translation; NMT; attention mechanisms; adaptive learning; context-aware translation; deep learning.
Abstract: Ambiguous sentences in machine translation can lead to inaccurate or contextually inappropriate translations, which is a remarkable challenge. In this paper, we introduce a translation system that has a context-aware feature, and it is based on deep learning using transformer-type neural networks with attention mechanisms to amplify disambiguation. In particular, our model uses contextual embeddings and syntax-semantic analyses for model training to ensure translation accuracy, especially in lexical, syntactic, and referential ambiguity cases. We test our system against highly regarded translation systems and show that our model is capable of meaning preservation and fluency improvement. The experimental results show a remarkable performance upgrade, especially in translating low-resource and idiomatic texts. This study demonstrates how deep learning dynamically tailors translation to context, improving disambiguation and fluency. Keywords: translation; deep learning; context-aware; ambiguity resolution; neural networks.
Abstract: Machine translation (MT) has undergone a remarkable transformation with the rise of deep learning methods, significantly improving translation accuracy and fluency. This paper examines how deep learning methodologies have influenced MT, particularly through the use of neural networks, Seq2Seq models, attention mechanisms, and transformer architectures. Traditional rule-based or statistical approaches have evolved into neural machine translation (NMT), leveraging large-scale data and advanced learning paradigms. The study highlights advancements, challenges, and future prospects, focusing on low-resource language translation, model bias, and computational efficiency. By analysing current developments and trends, this paper emphasises the revolutionary role of deep learning in enhancing multilingual communication through machine translation. Keywords: deep learning; machine translation; MT; neural networks; transformer models; language processing.
Abstract: Machine learning is changing the face of college English education with personalisation, immediate feedback, and student performance prediction. In this study, we propose an ML-driven framework. The first stage is an RNN attention model-based content recommendation, followed by a fine-tuned GPT-4 model for writing evaluation and an optimised random forest model for early risk detection. On real-world data, we achieve a 31.2% increase in recommendation accuracy, a 49% reduction in grading time with high BLEU and ROUGE scores, and 88.3% accuracy in identifying at-risk students. If data privacy, Ethics, and the like are appropriately managed, ML increases student performance by 27% and grading efficiency by 40%. Keywords: machine learning; personalised learning; natural language processing; NLP; automated feedback; predictive analytics; English language teaching.
Abstract: This article explores the innovative application of deep learning technology in reimagining ethnic elements in animation, based on Jungian archetype theory. Addressing the homogenisation of traditional cultural symbols in animation amid globalisation, a three-dimensional creation model of archetype decoding-intelligent generation-cultural verification is proposed. By building a deep neural network database of traditional patterns, mythological themes, and opera elements, and utilising generative adversarial networks (GANs) and variational autoencoders (VAEs), cultural archetypes are deconstructed and reassembled. Case studies demonstrate that this approach effectively extracts collective unconscious features from ethnic elements while preserving the spiritual core of cultural archetypes, generating innovative visual expressions with modern aesthetics. The research offers interdisciplinary insights for the innovative inheritance of cultural heritage from a digital humanities perspective and opens new technological pathways for animation creation in the AI era. Keywords: deep learning; generate adversarial networks; variational autoencoder; VAE; animation creation.
Abstract: Art style image migration has been increasingly important in image processing research as computer vision and deep learning technologies develop. Most well-known style migration techniques depend on paired training data, which is occasionally difficult to get in practice. Many solutions for complex art forms lose material or have style inconsistencies, which makes it challenging to fulfil high-quality content preservation. Regarding the mentioned problems, this work presents ArtCycleGAN, a cycle-consistent generative adversarial network-based art style image migration model. Pre-trained VGG19 network perceptual loss and cycle-consistent loss help to enable high quality unsupervised art style migration. ArtCycleGAN proves its validity in art style migration by experimental findings showing good performance in style similarity, content retention, and perceptual quality. This work presents a dependable and efficient approach for unsupervised art style migration as well as fresh ideas and references for picture synthesis applied with generative adversarial networks. Keywords: art style image migration; CycleGAN; perceptual loss; unsupervised learning; image generation; content retention.
Abstract: To enhance the sentiment correlation between images and texts, this paper proposes a multimodal sentiment analysis approach for image-enhanced text from a linguistic perspective (TSAIE-AppLing). Firstly, bidirectional encoder representations from transformers (BERT) are introduced to encode textual features, and image features are extracted using visual transformer, which is combined with a multi-head self-attention mechanism to capture cross-modal global semantic features. Then we use null convolution to strengthen the feature association between image blocks and aggregate cross-block features, design a multi-head cross-attention mechanism to achieve inter-modal interaction alignment, use graph convolutional network (GCN) to enhance the textual semantic features related to the image, and carry out the final sentiment polarity determination through softmax function. Experimental results on the MVSA dataset show that the proposed method improves the classification accuracy by at least 2.75%, which can significantly improve the multimodal sentiment analysis. Keywords: multimodal sentiment analysis; BERT model; multihead cross attention; graph convolutional network.
Abstract: At present, English speech teaching has developed into an intelligent form, and speech recognition function is combined in speech teaching to perform spoken English correction, but the accuracy of speech recognition needs to be improved. In order to improve the effect of speech intelligent recognition in English speech teaching, this paper combines speech recognition technology with visualisation technology to propose a speech recognition visualisation technology, designs and implements ASR algorithm based on Conformer encoder and CTC decoder, and realises VITS speech synthesis model. At the same time, this paper uses knowledge distillation method to obtain a lightweight speech synthesis model, uses MobileNetV3 network to realise the lightweight YOLOv5s model, and combines DeepSORT tracking algorithm and statistical function to realise the target statistical function. According to the comprehensive test results, it can be seen that the model proposed in this paper has high speech recognition accuracy and speed. In addition, it can be seen from the comparative test results that the model proposed in this paper has certain advantages in speech recognition compared with the existing research, and can meet the needs of intelligent English pronunciation teaching. Keywords: speech; visualisation; English; speech; teaching effectiveness. DOI: 10.1504/IJICT.2025.10071259
Abstract: This paper presents the design and implementation of an autonomous vehicle system for warehouse logistics, utilising a dual-vehicle coordination mechanism to enhance efficiency and adaptability in dynamic environments. The system integrates advanced hardware components, including the STM32 microcontroller, TB6612FNG motor driver, and YB-MVX01 infrared line tracking module, to achieve precise navigation and task execution. Wireless communication is facilitated through JDY-31 Bluetooth and GL24S 2.4G modules, enabling external control and inter-vehicle coordination respectively. A PID control algorithm ensures accurate line tracking, while an LCD display provides real-time system status. Experimental results demonstrate the systems effectiveness in autonomous navigation, task coordination, and adaptability to dynamic warehouse environments. The design prioritises modularity, scalability, and cost-efficiency, making it a viable solution for modern logistics automation. Keywords: microcontroller; autonomous vehicle system; dynamic warehouse environments; GL24S 2.4G modules. DOI: 10.1504/IJICT.2025.10071260
Abstract: Student management is a crucial aspect of educational institutions, encompassing activities such as admissions, academic tracking, attendance, and behavioural monitoring. Traditional management systems often lack the ability to generate actionable insights, which affects performance evaluation and decision-making. This study proposes a hybrid machine learning model that integrates supervised and unsupervised learning techniques to enhance student management. Using historical academic records, attendance logs, and behavioural data, the model predicts academic outcomes, identifies at-risk students, and suggests interventions. It employs clustering for student segmentation, predictive modelling for learning outcomes, and adaptive learning support for dynamic decision-making. Real-world data evaluations show improved accuracy and reliability over conventional approaches. The proposed model offers a scalable, intelligent solution for modern student management challenges. Keywords: student management; machine learning; ML; hybrid mode; academic performance; risk prediction.
Abstract: This study aims to analyse the students feedback data for enhancing the educational system. Teachers feedback serves as a critical tool for assessing educational outcomes and improving teaching strategies. Natural language processing (NLP), an active research area of artificial intelligence (AI), offers novel solutions for analysing and understanding large volumes of feedback data, aiding in the refinement of educational colleges. This paper aims to carry out a comprehensive analysis of students feedback by classifying content into five classes using advanced AI techniques including machine learning, ensemble methods, and deep learning (DL) combining with both textual features and word embedding features to improve predictive performance. Among all the applied features, the hybrid approach of the latest technique of FastText with DL model of Bi-GRU reveals the highest results with accuracy of 95%. This research confirms that NLP features provide deep insights into content and help us predict the various aspects of students feedback for improvements in the educational sector. Keywords: education; natural language processing; NLP; artificial intelligence; AI; sentiment analysis; deep learning; DL; feedback analysis.
Abstract: As the electricity trading market expands and becomes more complex, ensuring user safety and efficient equipment operation has become a critical challenge for the power industry. Inspection path planning and process modelling, as core technologies in intelligent inspection within the smart manufacturing system, have become essential tools for addressing this challenge. In response to the low efficiency of power system inspections, this paper proposes an intelligent inspection path optimisation and process modelling method (DDQN-GA) based on a combination of double deep Q-Network (DDQN) and genetic algorithm (GA). First, the proposed method employs the DDQN algorithm to intelligently allocate power-trading users and inspection teams, allowing each team to be optimally scheduled based on real-time system status and demand. Subsequently, GA is used to optimise the internal paths of each inspection team, effectively exploring and optimising complex path combinations to minimise overall inspection costs and achieve the optimal inspection plan. Experimental results demonstrate that this method significantly reduces total inspection costs and shortens computation time. Compared with three traditional algorithms, the DDQN-GA approach considerably improves computational efficiency, especially in handling large-scale inspection teams and user allocations. Keywords: power trading; user inspection; path planning; double deep Q-network; genetic algorithm; GA. DOI: 10.1504/IJICT.2025.10071013
Abstract: Traditional deep learning detectors often struggle to generalise when detecting diffusion-generated content. To address this, we propose DIRE, a generalised detector leveraging reconstruction error image representation. The framework standardises facial feature spaces through constrained feature learning and introduces a gradient suppression algorithm to filter abnormal gradients, preventing shortcut learning and enhancing generalisable feature extraction. Experiments on hybrid datasets validate DIREs effectiveness in four cross-domain tasks (O&C&M→I, O&C&I→M, O&I&M→C, and I&C&M→O). Ablation studies confirm the synergy of feature standardisation and gradient suppression, reducing bias by 97.6% and parameters by 42%, while accelerating inference by 2.3×. DIRE achieves 98.2% and 96.7% accuracy on two tasks (O&C&I→M and O&M&I→C), outperforming state-of-the-art methods by 5.3% while maintaining computational efficiency. This study advances generative face detection through dual optimisation, offering a lightweight framework for financial identity verification and social media content moderation. Keywords: denoising diffusion probabilistic models; DDPMs; diffusion reconstruction error; DIRE; face anti-counterfeiting; disentanglement. DOI: 10.1504/IJICT.2025.10071014
Abstract: This is one of the first regression machines that data scientists will encounter because it is easier and easier to interpret. It suffers from complex, nonlinear and high dimensional data. In the context of finance, healthcare and climate domains, this study suggests a hybrid machine learning framework combining Adam, RMSProp, XGBoost, SVMs, and neural networks to significantly improve the regression performance. MSE, R2 and efficiency metrics are used to analyse real world datasets. The financial forecasting MSE is reduced by 18% and the healthcare R2 improved by 22%. Noisy data was easier to deal with for climate models. Because they preserved interpretability, features were indicated by SHAP values. Blending classical statistics with modern AI transforms the problem into more accurate, scalable, and interpretable models, providing robust solutions for todays complex data challenges, which are proven. Keywords: AI-driven regression enhancement; ensemble learning regression; predictive modelling AI; SHAP interpretability in regression; gradient optimisation regression; and automated feature selection in regression. DOI: 10.1504/IJICT.2025.10071126
Abstract: Ensuring the quality of agricultural products in e-commerce is a significant challenge due to product variability and the absence of direct inspection before purchase. Customer reviews serve as a critical source of information, offering insights into product freshness, packaging, and overall satisfaction. This research focuses on the agricultural product domain, where quality plays a pivotal role in ensuring consumer trust and satisfaction. Harnessing the power of large language models (LLM), this study investigates the application of state-of-the-art transformer-based models for analysing customer feedback. The research utilises fine-tuned BERT and RoBERTa models to classify and predict product quality based on sentiment and contextual analysis of user reviews. The findings highlight the remarkable performance of these models, with RoBERTa achieving the highest accuracy of 99%. This study signifies the growing importance of AI and LLMs in enhancing e-commerce practices, particularly in domains like agriculture, where product quality assessment is paramount. Keywords: artificial intelligence; e-commerce; agriculture; transformers; deep learning; sentiment analysis; natural language processing; NLP; machine learning. DOI: 10.1504/IJICT.2025.10071015
Abstract: The health status of the young and middle-aged population has a significant impact on the stable functioning of society. To address the issue of poor prediction accuracy in the current study, the BP neural network (BPNN) is first improved based on Bayesian optimisation (BO), and the parameter combination that maximises the conditional probability is selected to improve the fitting accuracy of the model. Then the factors affecting the health status are analysed, the influencing factors are decomposed and reconstructed using the improved variational modal decomposition (VMD) and fuzzy entropy algorithm, and the objective function is iteratively searched through BO probability theory to obtain the degree of the BPNN parameter that minimises the prediction error. Finally, corresponding health management suggestions are proposed for the prediction results. The simulation results indicate that the Accuracy of the proposed method is 95.37%, which significantly improves the prediction accuracy. Keywords: health prediction; Bayesian optimisation; probability theory; BP neural network; BPNN; variational modal decomposition; VMD. DOI: 10.1504/IJICT.2025.10071016
Abstract: The image quality generated by the traditional virtual try-on technique for non-heritage accessories is poor, to address this problem, this paper firstly designs a convolutional neural network that adaptively adjusts the feature extraction strategy, and adopts an improved generative adversarial network to generate a primary feature map. Then the background noise of the primary feature map is suppressed based on multi-scale attention, and an adaptive perceptual enhancement module is designed to weight the features at different locations in the feature map to strengthen the representation of important features. Finally, the primary feature maps are perspective corrected and downscaled using multi-scale weights to enable the network to generate high-quality images of non-heritage accessory try-on. Experimental results on UNESCO and VITON datasets show that the structural similarity (SSIM) of the suggested method improves 3.45-26.76% compared to benchmark methods, which can effectively improve the quality of image generation. Keywords: virtual try-on; convolutional neural network; generative adversarial network; GAN; multiscale attention; adaptive perceptual enhancement. DOI: 10.1504/IJICT.2025.10071017 |