International Journal of Bio-Inspired Computation (33 papers in press)

Regular Issues

• Research on Philosophy Sentence Segmentation Algorithm Based on BERT Pre-training and Deep Learning Network  
  by Dan Cheng 
  Abstract: To improve the accuracy of sentence breaking in ancient philosophy texts, based on BERT pre-training and deep learning network, an ancient philosophy sentence segmentation algorithm with high accuracy is proposed. Based on Lattice model, BERT model is used to encode ancient philosophical texts into word vectors, and Flat-Lattice Transformer (FLT) position coding method is adopted to determine the interaction relationship between vectors. Then, sentence segmentation situation of ancient philosophical texts is predicted. Finally, sentence segmentation accuracy of ancient philosophical texts is improved. The simulation results show that compared with Lattice-LAN, BERT-Bi-LSTM-CRF and Bi-LSTM-CRF models, the proposed algorithm has obvious advantages in the accuracy of ancient philosophy sentence segmentation. Moreover, accuracy, recall, and F-value of the proposed algorithm are 82.82%, 70.71%, and 76.29%, respectively, which improves the sentence segmentation accuracy of ancient philosophical texts and has certain reference value.
  Keywords: philosophy sentence segmentation; BERT pre-training; deep learning; LSTM model; FLT position coding.
  DOI: 10.1504/IJBIC.2024.10068326
   
  
• Research on Sports Image Restoration Algorithm based on Deep Learning  
  by Xinjiao Zhang, Xuefei Tao 
  Abstract: To improve the sharpness and realism of sports image restoration, an image restoration method based on deep learning is proposed. The improved GAN (Generative Adversarial Network) is adopted for image denoising, and the improved RAiA-Net is used for image rain removal. The simulation results show that the peak signal-to-noise ratio and structural similarity of the improved GAN network reach 32.56 and 0.91, respectively, which are 3.45 and 0.04 higher than those of GAN network. The peak signal-to-noise ratio and structural similarity of RAiA-Net are 36.87 and 0.98 respectively, and the time consumption is shorter (0.43s), which has certain advantages and practicability. Compared with the standard GAN network, CBM3D network and CNN network, the designed network has better image denoising performance and better performance in peak signal-to-noise ratio and structural similarity indexes.
  Keywords: deep learning; image restoration; image denoising; image rain removal; generative adversarial network.
  DOI: 10.1504/IJBIC.2024.10068652
   
  
• Contrastive Hypergraph Attention Reinforcement Learning for Asset Portfolio Management in Digital Auditing  
  by Xuan Liu, Xu Cao, Xinying Fang, Nan Gao 
  Abstract: Asset portfolio management (APM) is a critical task within digital auditing, aimed at continuously reallocating funds across a set of assets with the aim of maximising investment returns or suppressing risks. Recent APM methods have learned a policy function to generate appropriate portfolios in a reinforcement learning framework and have achieved encouraging progress. However, these methods ignore the fact that assets in portfolios may be broadly interrelated and affected by each other, and the asset relationship information is highly valuable in improving APM. In this paper, we introduce group-wise relationship information as additional environmental cues to improve portfolio policy learning for APM. We propose a dual-channel hypergraph attention network to jointly capture the heterogeneous fund-holding and industry-belonging relationships via attention-based information aggregation among assets. In addition, we introduce contrastive learning to provide extra supervision signals for higher modelling capacity. Experimental results demonstrate that this method yields significantly greater benefits compared to state-of-the-art methods for APM.
  Keywords: Asset portfolio management; Hypergraph learning; Contrastive learning; Attention;Reinforcement learning.
  DOI: 10.1504/IJBIC.2024.10069679
   
  
• Financial Fraud Detection Based on Distributed L  
  by Hongwei Chen, Jianpeng Wang, Lun Chen, Zexi Chen, Rong Gao, Xia Li 
  Abstract: With the growth of online payments, financial fraud has led to significant economic losses. Accurate fraud detection is therefore crucial for protecting consumer rights and ensuring the security of payment platforms. Current detection methods mainly rely on machine learning algorithms, but their performance is often limited by hyperparameter tuning constraints. To address this, we propose a Spark-based distributed L
  Keywords: fraud detection; sailfish optimiser; Lévy flight; Spark.
  DOI: 10.1504/IJBIC.2025.10069838
   
  
• MOWOA-D based Feature Selection Study for Pneumonia Detection  
  by Leyi Xiao, Yixuan Su, Xia Xie, Chaodong Fan 
  Abstract: AI's application in medical image analysis presents challenges in balancing solution diversity and convergence in high-dimensional datasets for accurate feature selection. To this end, we propose an improved multi-objective evolutionary algorithm (MOWOA-D) based on decomposition and Whale Optimization Algorithm (WOA) learning strategy. This algorithm integrates the WOA learning strategy into the original MOEA/D to optimize the search mechanism, ensuring accuracy and uniformity. Then we design a dynamic neighborhood reference strategy to select reference points based on the population's real-time state, guiding the generation of offspring and enhancing solution quality. We also use an adaptive binary mutation operator that introduce the Hamming distance and adjust mutation probability during iteration to balance exploration and exploitation. Experimental results indicate that MOWOA-D outperforms five multi-objective evolutionary algorithms in efficiency and uniformity. Feature selection experiments on the UCI database show improved classification accuracy and reduced data redundancy. Tests on pneumonia data demonstrate the method's practicability.
  Keywords: Multi-objective optimization; Feature selection; MOEA/D; Pneumonia; Whale optimization; Heuristic method; Dynamic reference strategy; Binary mutation operator; Hamming distance; Machine learning.
  DOI: 10.1504/IJBIC.2025.10070157
   
  
• Student Performance Prediction (SPP)Model using HLRO-DMN  
  by L. Srinivasan, D. Kalaivani, C. Nalini, I. Gugan 
  Abstract: This research introduced the proposed hybrid leader remora optimisation algorithm with deep maxout network (HLRO_DMN) for accurately predicting the students performance. Initially, the input data acquired from the dataset is transformed into a suitable format using Yeo-Johnsons transformation. Then, the dice coefficient is employed for selecting optimal features, which combines the feature score obtained from the Fisher score and the Tversky index. In addition, the data augmentation is completed by the bootstrapping method, and the performance prediction is carried out by the DMN, wherein the weight of the DMN is tuned by the HLRO algorithm. Besides, the experimentation of HLRO_DMN attained the best result using certain metrics, like mean square error (MSE), Root mean square error (RMSE), and mean absolute error (MAE), and the accuracy of the corresponding values noted by the devised scheme are 5.4032, 0.175, 0.4444, and 91.314, respectively.
  Keywords: Remora Optimization algorithm; Deep Maxout network; Hybrid leader based optimization; Yeo-Johnson's transformation; Dice coefficient.
  DOI: 10.1504/IJBIC.2025.10071048
   
  
• Spatial and Temporal Distribution of Ecosystem Service Values in the Northeast Tiger and Leopard National Park and Analysis of Driving Factors based on the Geographical Detector and Remote Sensing Classification  
  by Zhihan Wan, Hongxun Li 
  Abstract: This study provides a basis for research on the value of ecosystem services of Northeast Tiger and Leopard National Park (NTLNP) after the implementation of the natural protection project by analyzing the change of land use, temporal and spatial changes of ecosystem service value, and driving factors from 2000 to 2022. The results showed that the area of broad-leaved forest in NTLNP was the largest and the land use types changed from broad-leaved forest to mixed coniferous and broad-leaved forest and coniferous forest from 2000 to 2022; the ecosystem service value of NTLNP showed a trend of first decreasing and then increasing, with a total increase of 464.99 million yuan. The interaction between elevation and climate factors has a great influence on the value of ecosystem services, and natural factors have a binding effect on the spatial differentiation of ecosystem service value.
  Keywords: Northeast Tiger and Leopard National Park; ecosystem services value; driving factors; geographical detector.
  DOI: 10.1504/IJBIC.2025.10071729
   
  
• Research on Regional Low-Carbon Economy Analysis Model based on Improved SVM  
  by Na Xiao, Yan Liu 
  Abstract: In order to further grasp the level of low-carbon economic development in the region and provide feasible suggestions for regional development, this paper proposes a regional low-carbon economy analysis and prediction model based on multiple strategies improved sparrow search algorithm optimised SVM (ISSA-SVM). Where, support vector machine (SVM) is used as the basic prediction method, and sparrow search optimisation algorithm improved based on multi strategy and combine with PLS component extraction is introduced to optimise the parameters of SVM and further enhance its predictive performance. Simulation results show that the designed prediction model based on ISSA-SVM has good prediction performance. Furthermore, the prediction error rate of this model has always been stable below 2.5%, and its precision is high. Therefore, the proposed model can be used to analyse and predict the development level of low-carbon economy in practice, which has certain feasibility.
  Keywords: low-carbon economy; support vector machine; sparrow search algorithm; prediction model.
  DOI: 10.1504/IJBIC.2025.10072382
   
  
• Classification of Malaria Disease using Optimal Fire Hawks based Capsule Convolutional Vision Transformer Network  
  by Pallavi Bhanudas Salunkhe, Pravin Sahebrao Patil 
  Abstract: Malaria is a widespread harmful disease caused by parasites transmitted through infected mosquitoes. The primary challenge addressed in this paper is the low performance, high error rates, and poor segmentation accuracy in malaria disease classification using existing machine learning (ML) and deep learning (DL) techniques. To overcome these limitations, this study proposes an advanced malaria classification model using the optimal fire Hawks-based capsule convolutional vision transformer network (OFH_C2ViTNet). The input image was segmented using feature concatenation-based U-Net (FC-UNet) based on red blood cells, white blood cells, and malaria parasites. Finally, the suggested optimal fire Hawks-based capsule convolutional vision transformer network (OFH_C2ViTNet) was used to categorise malaria into difficult, trophozoite, gametocyte, ring, and schizont kinds. From the input images, the proposed model provides an efficient result for identifying the disease at an early stage. Experimental evaluation demonstrates that the proposed model achieves superior performance, with an accuracy of 97.42% and a precision of 98.68%, outperforming existing state-of-the-art methods in malaria detection.
  Keywords: Median Filtering; Contrast Enhancement; Feature Concatenation; Hawks Optimizer; Vision Transformer Network; Malaria Parasites.
  DOI: 10.1504/IJBIC.2025.10072835
   
  
• Self-Improved Optimisation Model On Energy-Aware Resource Deployment Algorithm for Cloud Data Centres  
  by Prabha B, Kalangi Ruth Ramya, Charanjeet Singh 
  Abstract: The high processing demands of business, social, web, and scientific applications are driving a sharp rise in the demand for cloud computing. Most of the current cloud data centre resource management algorithms take CPU utilisation as the main consideration and increase CPU utilisation through virtual machine integration to reduce the energy consumption of cloud data centres. This work intends to propose a new cloud resource deployment model that includes three major aspects. Clustering is the initial phase, where clustering takes place by the Improved FCM-based clustering. Here, it is used to group the physical machines under the consideration of Energy and Distance. The process of deployment of the virtual machine is handled by Optimisation assisted Bi-GRU. Thereby, the proposed optimisation problem will be solved by Updated SSA with Baker Map Evaluation. The improved FCM with enhanced Jensen-Shannon distance has achieved a high accuracy value of 0.944.
  Keywords: Cloud Computing; Cloud Data Centers; Improved Fuzzy C Means based clustering; Resource Deployment; Updated Squirrel Search Algorithm with Baker Map Evaluation Algorithm.
  DOI: 10.1504/IJBIC.2025.10073095
   
  
• Modified African Vulture Optimisation Algorithm using Inertia Weight and Clerc-Kennedy Formula  
  by Saleh Altbawi, Saifulnizam Bin Abd. Khalid, Rayan Hamza Alsisi, Touqeer Ahmed Juman, Zeeshan Arfeen 
  Abstract: This study introduces the Modified African Vulture Optimization Algorithm (MAVOA) as an enhancement of the existing African Vulture Optimization algorithm (AVOA). MAVOA addresses several limitations of AVOA, including suboptimal solutions and premature convergence due to demographic homogeneity. The MAVOA involve incorporating the Clerc-Kennedy formula and adjusting the inertia weight, which leads to faster convergence and better-quality hyper parameter approximation. One notable feature of MAVOA is its flexibility, allowing it to dynamically regulate its search strategy to balance local exploitation and global exploration. Comparative evaluations against other algorithms demonstrate the superior performance of MAVOA. This study also showcases MAVOA's effectiveness through real-world engineering challenges, highlighting its potential for various optimization applications.
  Keywords: African Vulture Optimisation Algorithm; metaheuristic; Optimization; Benchmark; Economic Load Dispatch.
  DOI: 10.1504/IJBIC.2025.10074201
   
  
• Robust Adaptive Iterative Learning Control for Nonlinear Discrete-Time System in Fading Environments with Joint Multiplicative-Additive Effects  
  by Yunshan Wei, Chengxi Liang, Kai Wan, Xingfeng Cai, Xinru Liu 
  Abstract: This study investigates the robustness of discrete-time adaptive iterative learning control (AILC) under fading channels for nonlinear dynamical systems with multiplicative and additive channel noises. The output and the input fading channel are considered integral components of the system during the analysis process, and thus, the system is reconstructed. Each component of system outputs suffers different multiplicative noise. The framework jointly addresses stochastic multiplicative and additive randomness effects in signal propagation. The variable tracking targets and error dead zones is taken into account in AILC design. The robustness characteristics of the developed AILC methodology are systematically examined. The validity of the developed approach is confirmed through systematic numerical simulations.
  Keywords: Adaptive iterative learning control; Fading channels; Robustness; Discrete-time systems; Dead-zone.
  DOI: 10.1504/IJBIC.2025.10074473
   
  
• Research on Parallelisation Prediction for the Size of Industrial Economies based on FA-SSA Optimisation SVM  
  by Xiaofang Shi 
  Abstract: Aiming at the problems of slow prediction speed and low accuracy of industrial economy scale, this paper proposes a prediction model based on SVM (Support Vector Machine) and parallel framework. Firstly, SSA (Sparrow Search Algorithm) algorithm improved by elite inverse strategy and firefly disturbance strategy is adopted to optimize the kernel function parameter and penalty factor C of SVM model, and the optimal parameters of SVM model are obtained by using the MapReduce parallel computing framework, which improves the prediction speed and accuracy of industrial economic scale. The results reveal that the mean absolute error, mean square error and explained variance of the proposed method are 0.31, 0.17 and 0.85, respectively. Therefore, the proposed model can be applied to the actual industrial economy scale prediction.
  Keywords: size of industrial economies; SVM model; sparrow search algorithm; parallel framework.
  DOI: 10.1504/IJBIC.2025.10074479
   
  
• A Multisensory Direction Cues Integration Model based on Continuous Attractor Neural Networks  
  by Jinhan Yan, Naigong Yu 
  Abstract: Accurate heading perception is essential for spatial navigation. The mammalian brain integrates self-motion cues from multiple sensory modalities in a Bayesian manner to achieve more reliable spatial localisation, providing important inspiration for robot navigation. This paper proposes a multisensory integration model based on continuous attractor neural networks (CANNs) to fuse directional cues from visual optical flow and vestibular system. Prior to integration, we perform a decoupling analysis of optical flow induced by self-motion and design a simple method to estimate visual angular velocity quickly. Separate CANNs are then constructed to perform angular path integration for both visual and vestibular inputs, followed by multisensory integration. Experiments on simulation and real-world datasets demonstrate that the proposed model achieves accurate and reliable direction estimation while maintaining low computational cost and strong noise resistance, outperforming several advanced methods. This work provides a brain-inspired solution for efficient and reliable navigation.
  Keywords: Bio-inspired navigation; head direction cells; continuous attractor neural network; multisensory integration; angular velocity.
  DOI: 10.1504/IJBIC.2025.10074549
   
  
• Enhanced Water Body Segmentation Using Optimized DeepLabV3+ and YOLOv7 on High-Resolution Satellite Imagery  
  by A. Kalaiselvi, T. Jarin, P. Sreeja, Rajesh V 
  Abstract: Efficient segmentation of water bodies from satellite images is crucial for environmental monitoring, resource management, and flood prediction. Traditional machine learning methods face challenges such as manual feature extraction and complex spectral analysis. This work introduces an advanced model that integrates optimised DeepLabV3+ and YOLOv7 to enhance the segmentation of water bodies using high-resolution satellite imagery. The model employs DeepLabV3+ enhanced with an improved mantis search algorithm (IMWSA) to better delineate water boundaries. Subsequently, YOLOv7 assists in localising and classifying water regions. This combination harnesses the robust segmentation capabilities of DeepLabV3+ with the rapid detection potential of YOLOv7, resulting in improved accuracy and efficiency. Comparative analysis with two benchmark datasets demonstrates the models superiority, achieving accuracies of 99.5% and 99.02%. This optimised approach is pivotal for sustainable water management and disaster mitigation, providing a significant advancement in water body segmentation and monitoring
  Keywords: satellite images; segmentation of water bodies; Optimized DeepLabV3+; YOLOv7.
  DOI: 10.1504/IJBIC.2025.10074607
   
  
• RETIT: a Recursive symmetry Temporal Interactive Transformer Used for Compressed Video Action Recognition  
  by Jiyuan Wang, Huilan Luo, Chanjuan Wang, Ting Li 
  Abstract: The growing adoption of compressed video across diverse applications underscores the demand for efficient action recognition methods. Traditional RGB-based methods face limitations, especially because they depend heavily on computationally intensive optical flow for temporal analysis. We introduce the recursive symmetric temporal interaction transformer (RETIT), which leverages the transformer architecture to enhance global temporal interactions directly from compressed video data. RETIT employs recursive strategies to iteratively refine motion representations and integrates a specialised cross self-attention mechanism to enhance multi-scale spatio-temporal feature extraction. When evaluated on the K400, UCF101, and HMDB51 datasets, RETIT achieved top-1 accuracies of 72.1%, 97.6%, and 75.8%, respectively, surpassing existing state-of-the-art benchmarks. These results highlight RETITs ability to effectively leverage spatial and temporal modalities, advancing the state of action recognition in compressed video formats.
  Keywords: Compressed Video Analysis; Action Recognition; Transformer Architecture; Spatio-Temporal Features; Recursive Frame Interaction.
  DOI: 10.1504/IJBIC.2025.10074698
   
  
• Global-Local Perception GAN For Text-to-Images Synthesis  
  by Li Jianghua, Zhang Shouxin 
  Abstract: Text-to-image synthesis is a challenging cross-modal task with significant applications in multimodal artificial intelligence. While generative adversarial networks (GANs) have advanced text-conditioned image generation, existing methods suffer from two critical limitations: 1) semantic inconsistency caused by modality information mismatch; 2) insufficient fine-grained details in synthesised images. In order to solve these problems, we propose global-local perception generative adversarial network (GLP-GAN) a single-stage framework featuring three innovative components: deep information matching module (DIMM) that aligns textual and visual semantics through cross-modal matching loss, dynamic refinement module (DRM) that progressively enhances image details via adaptive weight updating, and conditional instance-batch normalisation (CIBN) that stabilises training by learning semantic-visual correlation patterns. Experimental results demonstrate state-of-the-art performance on CUB and COCO datasets, achieving 10.8% FID reduction on CUB and 22.3% improvement on COCO compared to baseline model. The IS shows 3.3% gain on CUB and 6.1% enhancement on COCO, and the results demonstrate the effectiveness of the proposed method.
  Keywords: Information matching; text-to-image generation; generative adversarial networks; multimodality; attention mechanism.
  DOI: 10.1504/IJBIC.2025.10074735
   
  
• Psychological Health Detection Based on Transformer Stress EEG Signal Recognition Model  
  by Hongying Zhang 
  Abstract: In response to the problem of emotion recognition implied in electroencephalogram signals, this study conducts psychological health testing based on electroencephalogram signal recognition. Firstly, position embedding is added to the electroencephalogram signals, and a Transformer-based electroencephalogram signal recognition model is constructed. Then, using the attention mechanism to improve the long short-term memory network, an emotion recognition model based on the bidirectional network is constructed. The results demonstrated that the proposed model had high recognition accuracy on both datasets, with 94.67% and 95.06% respectively, and recall rates of over 90%. The recognition accuracy (91.46%) and F1 score (90.77%) of the complete model were the highest. The proposed emotion recognition model had the highest recognition accuracy (98.14%), the smallest loss value (0.02), and the fastest decline rate. The research results contribute to the development of psychological health monitoring technology.
  Keywords: Transformer; EEG signals; Psychological health; Multi-head attention; Long short-term memory.
  DOI: 10.1504/IJBIC.2025.10075039
   
  
• A Novel MSAO-F-SVM Classifier based on Feature Transformation and Parameter Optimisation  
  by Qian Wang, Qinghua Gu, Yan Wang 
  Abstract: The generalisation error performance of the support vector machine algorithm is affected by the radius-to-interval ratio. At the same time, the selection of model parameters largely determines the classification performance. However, the existing optimisation algorithms seldom take these two aspects into account. This article proposed a new radius-margin-based SVM model with improved Aquila optimizer called MSAO-F-SVM, which considers the maximisation of margin and the minimisation of radius information. Firstly, the AO algorithm is integrated with Tent chaos mapping, differential evolution algorithm, and the introduction of the adaptive weighting factor to search the optimal global parameters effectively. Secondly, the enhanced AO method is utilised to choose the ideal values for maximum performance. Finally, the model is solved in three steps: matrix initialisation, parameter optimisation and transformation matrix solution. Experimental results showed that the proposed MSAO-F-SVM algorithm has better classification accuracy compared to other models and is valuable for solving classification problems.
  Keywords: Support vector machine (SVM); Aquila optimizer (AO); Parameters optimization; Feature transformation; Algorithm improvement.
  DOI: 10.1504/IJBIC.2025.10075741
   
  
• Swarm Intelligence and its Application: an Overview  
  by Qianjin Guo, Hongye Wang, Shuqi Huangfu, Shuai Guo, WanRu Gao, Yazhou Hu 
  Abstract: Swarm intelligence is a fascinating concept inspired by decentralised and self-organised biological groups. It is valued for redundancy, robustness, diversity, and effective spatial coverage. This review examines its core concepts, methodologies, military applications, and challenges. Furthermore, we propose an innovative classification framework combining application scenarios and algorithm characteristics. Specifically, we overview the fundamental concepts, development, and working principles underlying swarm intelligence. The three essential elements and five key principles defining this approach are summarised. Then, we explore mainstream swarm intelligence algorithms, their applications, and improved iterations. A captivating aspect of swarm intelligence is its diverse military applications across aerospace, aviation, land, surface, and underwater domains. We elaborate on its use in these five fields, discussing current development and research progress. Finally, the major challenges swarm intelligence faces are analysed. In summary, this review provides a comprehensive exploration of swarm intelligence its principles, algorithms, military applications, challenges, and future prospects.
  Keywords: Intelligence swarm system; Swarm intelligence; Robot swarm; Multi-agent reinforcement learning; Intelligence algorithm.
  DOI: 10.1504/IJBIC.2025.10075743
   
  
• Reconfigurable FIR Filter Design Optimised with Hybrid Adolescent Identity Search Algorithm and Group Teaching Algorithm in FPGA for Biomedical application  
  by Senthil Kumar S, S. Karthick, S. Thillaikkarasi, Rajesh Kumar T 
  Abstract: Finite Impulse Response (FIR) filters are crucial in biomedical signal processing, providing precise filtering to extract relevant information from noisy signals. However, designing and implementing FIR filters for real-time applications on Field-Programmable Gate Arrays (FPGAs) presents challenges in performance optimisation and execution time reduction. To address these challenges, design and implementation of a reconfigurable FIR filter optimized with a hybrid Adolescent Identity Search Algorithm and Group Teaching Algorithm (Hyb-AISA-GTA) on FPGA for biomedical applications called RFIR-Hyb-AISA-GTA is proposed. The RFIR filter utilizes a Truncation and rounding-based scaling rounding approximation Multiplier (TOSAM) and the Error Reduction Carrying Prediction Approximate Adder (ERCPAA) for enhanced performance. Optimal filter coefficients are estimated using Hyb-AISA-GTA algorithm. The RFIR-AISA-GTA filter achieves lower stop band attenuation of 0.2340 dB and shorter execution time of 2.46seconds compared to existing methods. Tested on Virtex FPGA using Verilog in Xilinx ISE14.5, the filter effectively removes noise from ECG signals.
  Keywords: Biomedical application; Group Teaching Algorithm; hybrid Adolescent Identity Search Algorithm; Reconfigurable finite impulse response (RFIR) filter; maximum ripples.
  DOI: 10.1504/IJBIC.2025.10075752
   
  
• Talent Cultivation Method for Building Informatisation Based on Distributed Teaching Assistance Model  
  by Fang Wang 
  Abstract: This study proposes a teaching assistance model for the cultivation of building informatisation talents based on a distributed architecture. Through the integration of microservice architecture and cloud computing technology, the dynamic scheduling and efficient management of teaching resources are achieved. This model takes modular design as the core, combines key technologies such as service registration discovery and API gateway, and supports high concurrent access and elastic scalability. Experiments showed that the model significantly improved the prediction accuracy by introducing preference factors and attendance behaviour analysis in the performance prediction algorithm. Meanwhile, the system could still maintain a stable response speed in high-concurrency scenarios, and the delay was controlled within an acceptable range. The proposed distributed teaching assistance model is scalable, flexible, and highly available, providing theoretical and practical references for the cultivation of talents in building informatisation.
  Keywords: Distributed architecture; Microservice architecture; Teaching assistance model; Talents in architectural informatisation; Cultivation.
  DOI: 10.1504/IJBIC.2025.10075945
   
  
• Siamese Network Based on Self-Attention Dropout and Blur Pooling for Object tracking  
  by Ciyuan Wang, Jia Zhang, Zhiheng Wang, Xiaoyu Sun, Juan Wang 
  Abstract: Addressing the challenges faced by Siamese network-based object tracking algorithms, such as difficulties in effectively extracting salient features of the target due to the shallow architecture of AlexNet, and positional offsets caused by pooling downsampling. This paper first designs a saliency attention module inspired by the working principle of the biological visual system, and adds a saliency attention layer in the offline training process to improve the network's ability to learn significant features through adversarial learning. Then, a feature extraction network with Max Blur Pooling as the downsampling layer is constructed to reduce the aliasing effect during the downsampling process. Finally, the trained tracking network is tested and evaluated using the tracking benchmark dataset OTB100, and the saliency attention layer is not used in the online tracking test. The experimental results show that the comprehensive success rate and precision rate of the proposed ADB-SiamFC are 4% and 3% higher than those of the original SiamFC respectively, and the adaptability and robustness are significantly improved.
  Keywords: Object Tracking; Siamese Network; Visual Inspiration; Saliency Attention.
  DOI: 10.1504/IJBIC.2025.10076636
   
  
• RLaGA: Reinforcement Learning-Assisted Genetic Algorithm for Satellite Resource Scheduling  
  by Zhang Zhe, Chengyu Hu, Xuesong Yan, Wenyin Gong, Dongcheng Li 
  Abstract: Satellite communication is critical in modern information transmission, necessitating efficient resource scheduling to maximise bandwidth utilisation and transmission speed. To address the challenges posed by increasing scheduling difficulty under varying task scales, we develop a mathematical model for satellite communication resource scheduling that incorporates task density effects, resource constraints, and multi-objective optimisation. Specifically, the objective function aims to maximise the total weighted priority gain of scheduled tasks and the number of successfully completed tasks, using a weighted sum formulation. Furthermore, we propose a novel reinforcement learning-assisted genetic algorithm (RLaGA). By integrating Q-learning into the genetic algorithm, RLaGA dynamically adjusts crossover and mutation operations, significantly improving solution quality and accelerating convergence. Experimental results conducted across multiple task scales to simulate increasing scheduling complexity demonstrate that our approach outperforms traditional heuristic algorithms, delivering superior performance in large scale, high density scheduling scenarios.
  Keywords: Satellite communication; reinforcement learning; genetic algorithm; large-scale; scheduling difficulty.
  DOI: 10.1504/IJBIC.2025.10076712
   
  
• Mechanical Property Prediction of Non-oriented Electrical Steel using Machine Learning Methods  
  by Wangya Huang, Chaoyu Fan, Qi Deng, Qi Kang 
  Abstract: This paper combines fundamental theoretical knowledge in materials science with practical production experience to successfully develop an ensemble learning-based prediction model for the mechanical properties of non-oriented electrical steel, leveraging various machine learning methods such as Bootstrap Forest, K-Nearest Neighbors, Neural Networks, and Stepwise Regression. The model evaluation reveals an R-squared value of 0.978 or higher, indicating its robust performance. By successfully integrating this prediction model into the Manufacturing Execution System (MES), monitoring and prediction of mechanical properties have been achieved without the necessity of additional mechanical testing equipment.
  Keywords: Non-oriented Electrical Steel; Mechanical Property Prediction; Machine Learning.
  DOI: 10.1504/IJBIC.2025.10077197
   
  
• An explorative  
  by Qiong Wang, Jiahang Li 
  Abstract: Differential evolution (DE) is a powerful metaheuristic, yet it often struggles to balance exploration and exploitation, leading to premature convergence. To address this, this paper proposes CGODE, a novel DE variant combining a hybrid mutation operator with a modified opposition-based learning (OBL) strategy. The approach introduces three key contributions: a population diversity metric to assess search status, a mutation operator utilising Cauchy and Gaussian distributions to balance the search process, and an elite subpopulation OBL to accelerate convergence. CGODE was validated using 41 benchmark functions from the CEC 2017 and 2022 suites and two real-world problems. Results demonstrate that CGODE significantly outperforms existing DE variants, OBL methods, and state-of-the-art algorithms in over 65% of the tested functions, effectively solving the exploration-exploitation equilibrium problem.
  Keywords: Differential evolution; Population diversity; Exploration-exploitation; Opposition-based learning; Parameter control.
  DOI: 10.1504/IJBIC.2025.10077433
   
  
• Chronic Kidney Disease Diagnosis using Optimised Adaptive Auto Regressive Deep Recurrent Neural Network Model  
  by Karuppuchamy V, Palanivel Rajan S 
  Abstract: Chronic kidney disease (CKD) is a growing public health concern due to its silent progression and severe long-term complications. To address the limitations of existing diagnostic methods, this study proposes a novel optimised adaptive auto regressive deep recurrent neural network (Op_A2DRNN) model for early and accurate CKD detection. A hybrid Osprey hunger game optimisation (HOGO) method selects key features, while the enhanced gazelle optimisation algorithm (EGOA) tunes model parameters to reduce over-fitting and enhance convergence. The integration of adaptive auto regression with deep RNN improves training efficiency and learning capability. Experimental results demonstrate superior performance, achieving 98.2% accuracy and outperforming existing models like CNN-GRU and DBN.
  Keywords: Cholesky triangle; missing data; standard deviation; threshold parameters; hunger weight; local optima; standardization; autoregressive effects; elite matrix.
  DOI: 10.1504/IJBIC.2025.10077453
   
  
• ECG Reconstruction with Compressive Sensing and Enhanced Step Size Optimised Sparsity Adaptive Matching Pursuit Algorithm  
  by S. Karthikeyani, S. Sasipriya, M. Ramkumar 
  Abstract: Electrocardiogram (ECG) signal monitoring plays a critical role in the early detection and diagnosis of cardiovascular diseases. However, efficient transmission and storage of ECG data remain challenging due to the large volume of signals generated. To address these challenges, this work presents an effective compression-detection-based reconstruction method. The incoming ECG signals are first represented as sparse signals and then compressed. To improve compression efficiency, an optimisation strategy is proposed using the improved intelligent satin bowerbird optimiser (I2SBO). The original signal is subsequently reconstructed on the medical server using the observation matrix and an enhanced step size optimised sparsity adaptive matching pursuit (ESSO_SAMP) algorithm. The proposed method is validated using the MIT-BIH atrial fibrillation (AF) database. The model achieves a MSE of 495.52, and a reconstruction probability of 0.944. These results demonstrate the effectiveness of the proposed method in enhancing ECG signal reconstruction while maintaining a high compression ratio.
  Keywords: Observation matrix; restricted isometry property; time index; recovering signal; elitism; convergence; termination; judgement; stage index.
  DOI: 10.1504/IJBIC.2025.10077454
   
  
• The Application of Adaptive Residual Module Optimisation Transformer Model in Sports Training Human Pose Estimation  
  by Jinchi Yu, Xiaomin Gu 
  Abstract: To make more intelligent and objective comparisons of body posture in sports training, this study proposes a network model of adaptive residual module optimised Transformer. Existing methods often struggle with feature degradation in complex motion sequences and individual biomechanical variations during cross-subject analysis. The model optimises the information transfer process by using the residual transfer pathway between attention layers, effectively addressing gradient dissipation while preserving detailed kinematic features, while improving the comparison accuracy. In addition, the study employs a multi-view pose estimation method to collect and process data from key points of the human 3D skeleton. It also standardises and normalises their positional data through anthropometric scaling to improve the accuracy of motion posture analysis. The test results revealed that the improved model’s comparison accuracy is 7.3% higher than the standard Transformer, with notable advantages in self-occluded postures and fast transitional movements, boosting intelligent sports development.
  Keywords: Residuals; Transformer model; Posture; Comparison; Attention layer; Sports training.
  DOI: 10.1504/IJBIC.2025.10077455
   
  
• Recognition of Chinese sentence intention combining multi-channel attention convolution and graph neural network  
  by Xin Zhao, Qiansong Wang 
  Abstract: To improve the recognition accuracy for Chinese sentence intention, this paper proposes a Chinese sentence intention recognition model by combining multi-channel attention convolution and graph neural network. The semantic features of Chinese sentences are extracted by using the improved CNN network, and the syntactic relations of Chinese sentences are extracted by using the graph neural network. Finally, the semantic features and syntactic relations are fused and softmax classifier is used for classification. The results show that average precision, recall and F1 value of the proposed model for different Chinese sentence intention recognition tasks reach 93.52%, 93.87% and 93.06%. Thus, the proposed model can improve the Chinese sentence intention recognition precision.
  Keywords: intention recognition; CNN network; graph neural network; semantic feature; syntactic relation.
  DOI: 10.1504/IJBIC.2025.10077456
   
  
• SExpSTO-ResNet: Serial Exponential Siberian Tiger Optimisation Algorithm-based ResNet50 for heart disease prediction  
  by Chitra M. G, Ramya Govindaraj 
  Abstract: This work designs a technique for heart disease prediction by deep learning model. The pre-processing stage is primary phase, and it is considered as most significant process to enhance the technique's performance. The main aim of pre-processing is to transfer raw data into processable data. Moreover, missing data imputation is exploited to carry out pre-processing that is employed to eradicate infinity values for effectual processing. Subsequently, by using the temporal convolutional network, feature fusion is done. Finally, heart disease prediction is done by using ResNet 50 that is trained by proposed serial exponential Siberian tiger optimisation named (SExpSTO) scheme that is derived by combining serial exponential weighted moving average in Siberian tiger optimisation (STO). The performance analysis of proposed algorithm is executed by considering parameters, like accuracy, sensitivity, and specificity. Finally, the experimentation evaluation is performed with maximum accuracy of 95%, maximum sensitivity of 97%, and maximum specificity of 94%.
  Keywords: Healthcare; Heart disease; feature fusion; deep learning; optimization algorithm.
  DOI: 10.1504/IJBIC.2026.10077626
   
  
• A Study on the Extraction of Topic Keywords from English Test Questions Text Based on Genetic Algorithm  
  by Yixian Lyu 
  Abstract: In order to improve the efficiency and accuracy of topic keyword extraction,a genetic algorithm combining word frequency and inverse document frequency is studied,and a weight evaluation model is constructed using multi-threaded computation and text processing to improve algorithm efficiency.Comparing the performance of different algorithms,it was found that the current research algorithm has an accuracy of 98.9%and a runtime of 0.038 s,while the deep learning method has an accuracy of97%and a runtime of 0.085s.Therefore,the research results have demonstrated that the topic word extraction method based on genetic algorithm for English composition texts not only improves the computational accuracy of topic word extraction,but also has good application effects on word frequency extraction in English test questions,providing rich text vocabulary for vocabulary analysis in English test questions and learning platforms,and thus has good application potential in the field of English test questions.
  Keywords: English test questions; Topic keyword extraction; Genetic algorithm; Feature item weights; Text; Similarity.
  DOI: 10.1504/IJBIC.2026.10077627
   
  
• Oversampling Classification of Multi-class Imbalanced Data based on Metaheuristic Algorithm  
  by Xin-ji Chen, Jing-Lun Zhu, Ze-yu Liu, Jian-wei Liu 
  Abstract: Classifying imbalanced data is one of the most critical tasks faced in modern data analysis. Particularly, when combined with other factors such as the presence of noise, class overlap issues, and ambiguities, data imbalance significantly impacts classification performance. Furthermore, with deeper investigations into the class imbalance problem, it’s found that multi-class imbalanced learning is more common in practice. Despite this, current research in the field of data imbalance largely focuses on binary classification problems, with relatively less study on the more challenging multiclass classification issues. In this paper, we introduce a novel oversampling technique the multi-class sine cosine evolutionary hybrid algorithm (MSCEHA). This method leverages the global search capability of meta-heuristic algorithms and the genetic characteristics of differential evolution algorithms to generate samples with diversity and robustness, and it is less affected by outliers compared to existing oversampling methods. Lastly, by introducing the triangular decomposition strategy for handling multi-class problems, MSCEHA is less affected by the loss of inter-class relationship information than traditional multi-class decomposition strategies. Experimental results on multi-class imbalanced benchmark datasets show that our method has higher robustness to noise and performs better compared to existing methods.
  Keywords: Metaheuristic Algorithms; Class Imbalance; Oversampling; Decomposition Strategy; Multi-class Data.
  DOI: 10.1504/IJBIC.2025.10077635