International Journal of Computing Science and Mathematics (9 papers in press)

Regular Issues

• Construction of Course Arrangement System for College Sports Teaching Based on Adaptive Factor Optimised SGA  
  by Chunyu Yin, Li Shi 
  Abstract: The course arrangement method for college sports teaching has problems of low efficiency and low quality of course arrangement. This study proposes an intelligent course arrangement method based on improved dual population genetic algorithm. Firstly, course arrangement problems are described, and mathematical model is built. Then, the lowest running time, the highest teacher utilization rate and the best class distribution balance are taken as optimization objectives. Afterwards, an improved dual population genetic algorithm is used to solve course arrangement problems of physical education teaching. Finally, this algorithm is applied to the course arrangement system of college sports teaching for testing and analysis. Experimental results show that the proposed improved dual population genetic algorithm can solve course arrangement problems accurately and quickly. Furthermore, its average running time, average classroom utilisation rate and average class time distribution balance are 53.59s, 96.42% and 0.9, respectively, which are better than those of intelligent course arrangement methods based on ant colony algorithm, firefly algorithm, particle swarm optimisation algorithm, and grey wolf optimization algorithm.
  Keywords: adaptive factor; dual population genetic algorithm; competition mechanism; college sports teaching; intelligent course arrangement.
  DOI: 10.1504/IJCSM.2025.10070151
   
  
• A Novel Superpixel Segmentation Method for Improved Image Segmentation using Soft Probabilities and Local Pixel Variation  
  by Ke Liu 
  Abstract: Superpixel segmentation is a critical technique in computer vision, widely used in tasks such as image segmentation and object recognition. However, traditional methods struggle with complex and irregular boundary regions, often resulting in inaccurate boundary delineation and irregular superpixel shapes. To address these challenges, this paper proposes a novel superpixel segmentation method that introduces soft probabilities and leverages local pixel variation, combined with a weight function derived from edge detection to enhance global contextual information. Soft probabilities allow smoother transitions between pixel regions, while local pixel variation helps capture fine-grained details in boundary regions. Evaluations on the improved BSDS500 dataset show a 4% region similarity improvement, 8% information entropy reduction, and moderate accuracy gains over the best current method. The results highlight the method’s effectiveness in managing complex boundary information while maintaining high segmentation accuracy, representing a significant advancement in superpixel segmentation.
  Keywords: Superpixel segmentation; Edge detection; Soft probabilities; Image boundary segmentation; Local pixel variation.
  DOI: 10.1504/IJCSM.2025.10070863
   
  
• Monthly Product Sales Forecast based on Hybrid Prediction Models  
  by Weidong Lou, Yong Jin, Hailong Lu, Yanghua Gao 
  Abstract: To address the poor performance of existing methods in predicting monthly product sales and further improve the prediction accuracy, this paper selected three individual prediction models: ARIMA, linear regression, and SVR. Dynamic weighting factors, determined through grid search, were used to construct a hybrid prediction model. Taking cigarette products as an example, the three models were trained on the monthly sales data of various cigarette grades from 2019 to 2022, and then used to predict the sales data for 2023. A hybrid prediction model was constructed by applying grid search to find the optimal dynamic weighting factors for the three models, fully utilising the advantages of each. Compared to individual prediction models, the hybrid prediction model produced a smaller absolute percentage error, with outputs closer to actual data, making it more in consistent with real-world conditions.
  Keywords: Monthly product sales; Hybrid prediction model; ARIMA model; Linear regression model; SVR model.
  DOI: 10.1504/IJCSM.2025.10071000
   
  
• A Multi-Objective Optimisation Model for Mutual Matching of Machine Capacity and Product Specification based on Knowledge Graph and Inventory-to-Sales Ratio  
  by Qi Sun, Huihui Gao, Yonglong Yu, Wei Jian, Liuyan Jiang, Wanjiang Wang, Yikai Han, Renwang LI 
  Abstract: This paper takes the production scheduling of industrial enterprises in the tobacco industry as an entry point, analyses the production scheduling process of H cigarette factory, and takes the inventory-to-sales ratio scheduling strategy in the production scheduling process as the main object of analysis. With the goal of meeting market demand and reducing the inventory of finished cigarettes, a mathematical model of production scheduling according to the inventory-to-sales ratio is constructed, and the maximum and minimum inventory-to-sales ratios can be calculated. Combined with the knowledge graph model, a reasonable production scheduling strategy for the inventory-sales ratio is formulated. Combined with the annual inventory and sales ratio data of the tobacco industry and the capacity of the enterprise's wrapping machine, the model is optimised to reduce the number of inventories and reasonably use the capacity of the wrapping machine, so as to make the annual production quantity relatively balanced.
  Keywords: Inventory-to-sales ratio scheduling strategy; Knowledge graph; Packing machine capacity; Inventory.
  DOI: 10.1504/IJCSM.2025.10071001
   
  
• DCCA: 3D Object Detection with Dynamic Convolution and Channel Attention  
  by Chen Jiang, Shuxia Lu, Tingting Ma, Xianghu Zhou 
  Abstract: 3D object detection as a very important part of the autonomous driving perception obtained a rapid development in recent years, but the existing 3D detector depends on the fixed weight nuclear convolution to deal with the information area, and then aggregate context to detect, these methods in the relationship between the capture point, lose some information The dynamic selection process is more suitable for sparse and unordered point clouds data and can better extract features and aggregate context information We propose the Dynamic Convolution and Channel Attention module(DCCA) to enhance the existing 3D detector The attentional mechanism can be regarded as a dynamic selection process, which is realised by adaptive weights of features according to the importance of inputs. And our method can be flexibly applied to most of the most advanced detectors, improving the accuracy.
  Keywords: Autonomous Vehicle; Dynamic Convolution; LIDAR; 3D Object Detection.
  DOI: 10.1504/IJCSM.2025.10071465
   
  
• Bearing Fault Prediction Fusion Algorithm based on Multi-Dimensional Signal  
  by Lida Liu, Qimiao Wang, Mei Sun, Yingjie Chen, Peiguang Lin 
  Abstract: In practical industrial applications of fault prediction, due to the complex noise environment, the effective characteristics of extraction from the collected bearing vibration signal are very difficult and there is a problem of data breach. The prediction accuracy and adaptability of traditional models are insufficient. Therefore, this paper proposes a multi-dimensional signal-based CNN integration learning model. By optimising the pre-processing process, this model introduces methods including multi-label data balance and time frequency domain conversion; Build the optimal framework of the base model; use the CNN and the residual network integration learning algorithm fusion model to improve the accuracy and stability of the prediction. The experimental results show that the percentage of the model error is 20.52%, and as a result, DAF-LSTM (improved deep forest length memory neural network), CNN-LSTM (Space convolution long short-term memory neural network), ResNet-LSTM (long-time residual networks) were reduced by 4.66%, 7.72%, and 2.24%, respectively. Excellent performance can effectively improve the fault prediction capabilities of the bearings and ensure the safe and stable operation of industrial production.
  Keywords: Fault prediction,Deep learning; Frequency conversion; Convolutional neural network; Residual connection; Integrated network.
  DOI: 10.1504/IJCSM.2025.10071469
   
  
• The Study of the Adaptive Combination Optimisation Problem of the Handling Unit  
  by Zhang Mingyuan, Meng Wenjun, Sun Zhengyu, Zhao Xiaoxia, Sun Xiaoxia 
  Abstract: This study proposes an adaptive configuration search method to address inefficient formation selection and slow computation in multi-unit collaborative handling of special-shaped workpieces. The method introduces two evaluation metrics (fitness degree and remaining effective space) and two constraint strategies: an overstepping boundary constraint for extended placement possibilities, and a combination constraint for stability. Finally, these strategies are embedded into the genetic algorithm to propose a hybrid heuristic algorithm based on effective area. Validation on a dedicated dataset of special shaped workpieces demonstrates significant improvements the overstepping boundary constraint strategy reduces the total handling unit requirement by 23.94%, while the fitness degree and remaining effective space metrics achieve a 22.86% reduction in total unit demand while simultaneously significantly improving remaining space quality.
  Keywords: Intelligent logistics; Handling unit; Irregular workpiece; Combinatorial optimisation; Effective area.
  DOI: 10.1504/IJCSM.2025.10071601
   
  
• A Constrained Multi-Objective Optimisation with Weak-Side Complementary and Dynamically Guided  
  by Ziqiong Liu, Sanfeng Chen, Xiang Du, Wei Li, Hui Wang 
  Abstract: This paper proposes the SE-YOLOv5s network model for road tunnel detection, focusing on structural disease detection using a multi-sensor fusion approach. The model fuses multivariate heterogeneous data from multiline LIDAR and monocular cameras. Three key improvements are made to the YOLOv5 network: 1) the SE attention module is integrated into the YOLOv5s backbone to enhance feature extraction while reducing computational cost; 2) the DC-BiFPN feature fusion network replaces the PANet to improve small target detection; and 3) the EIoU loss function is used instead of CIoU to improve detection performance. Experimental results show that the improved SE-YOLOv5s outperforms YOLOv5s, YOLOv3, and YOLOv4, with Map@0.5 increases of 8.1%, 15.1%, and 10.1%, respectively. The results demonstrate the higher accuracy of the SE-YOLOv5s in detecting structural diseases in road tunnels.
  Keywords: Constrained multi-objective; Co-evolution; Evolutionary algorithms; weak-side complementary.
  DOI: 10.1504/IJCSM.2025.10071603
   
  
• A Road Tunnel Detection Method based on SE- YOLOv5 Network  
  by Yu Bai, Jichao Wang, Xuewei Zhang, Chaojie Zhang, Huiyuan Liu, Kongyun Chen, Jian Wang 
  Abstract: At the current stage, many complex optimisation problems can be transformed into constrained multi-objective optimisation problems (CMOPs). Constrained multi-objective evolutionary algorithms (CMOEAs) have become an efficient way of resolving constrained multi-objective problems. However, CMOPs will face huge challenges such as complexity of constraints, difficulty in exploring, and serious conflicts between objective functions and constraints. This paper proposes an efficient weak-side complementary and dynamically guided CMOEA named WDCMO. WDCMO has two populations: the main population and the auxiliary population. Its evolutionary process is divided into two stages: in the first stage, the WDCMO main and auxiliary populations focus on the exploration of two different regions; in the second stage, the WDCMO auxiliary population uses the information about the value of the objective function to guide the main populations evolution. Finally, WDCMO was tested against three other algorithms on a test suite. The experimental results show that the test values of WDCMO are clearly better than the other comparison algorithms on the majority of test problems. Specifically, WDCMO achieved 9 HV indicator leads and 12 IGD indicator leads on 14 constrained multi-objective problems.
  Keywords: Structural Disease Detection of Highway Tunnels; YOLOv5; Object Detection; Squeeze-and-Excitation Attention Mechanism.
  DOI: 10.1504/IJCSM.2025.10071604