Pub Date : 2025-05-01DOI: 10.1016/j.sasc.2025.200264
Xiangyu Wang , Kang Cao
The advent of cloud computing technology (CCT) has expedited the advancement of online learning methodologies and, to a certain extent, compensated for the limitations inherent in traditional teaching approaches. However, online teaching under CCT still has the problem of unstable teaching quality, so the study establishes a relevant learning effect evaluation model for the personalized learning platform of vocational education under CCT. To achieve more efficient and accurate evaluation of learning effect, an adjustable variation genetic algorithm-backpropagation neural network (AGA-BP) is proposed. The model introduces an adjustable mutation approach, which adapts the mutation probability in real-time in accordance with the progress of the genetic algorithm in the search process, so as to prevent entering into local optimization and ensure the maintenance of diversity. This strategy significantly enhances the convergence speed and overall search capability of the algorithm. Meanwhile, using the excellent fitting characteristics of neural network, AGA-BP model can accurately learn and simulate different students' learning behavior and effectiveness. The experiment outcomes indicate that the model's mean square error is 3.3883e*10–12, its fitness value is 1.36, and its average accuracy is 98.35 %.
{"title":"Personalized learning effect evaluation model for vocational education with cloud computing technology","authors":"Xiangyu Wang , Kang Cao","doi":"10.1016/j.sasc.2025.200264","DOIUrl":"10.1016/j.sasc.2025.200264","url":null,"abstract":"<div><div>The advent of cloud computing technology (CCT) has expedited the advancement of online learning methodologies and, to a certain extent, compensated for the limitations inherent in traditional teaching approaches. However, online teaching under CCT still has the problem of unstable teaching quality, so the study establishes a relevant learning effect evaluation model for the personalized learning platform of vocational education under CCT. To achieve more efficient and accurate evaluation of learning effect, an adjustable variation genetic algorithm-backpropagation neural network (AGA-BP) is proposed. The model introduces an adjustable mutation approach, which adapts the mutation probability in real-time in accordance with the progress of the genetic algorithm in the search process, so as to prevent entering into local optimization and ensure the maintenance of diversity. This strategy significantly enhances the convergence speed and overall search capability of the algorithm. Meanwhile, using the excellent fitting characteristics of neural network, AGA-BP model can accurately learn and simulate different students' learning behavior and effectiveness. The experiment outcomes indicate that the model's mean square error is 3.3883e*10<sup>–12</sup>, its fitness value is 1.36, and its average accuracy is 98.35 %.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200264"},"PeriodicalIF":0.0,"publicationDate":"2025-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143929343","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-01DOI: 10.1016/j.sasc.2025.200255
Xinxin Yang
Educational managers need curriculum evaluation results and academic warnings to enrich educational management content. The study integrates data analysis technology into education and teaching, and uses association rules to mine the internal relationship of each dimension element of course evaluation. The results show that the performance of the Apriori algorithm with interest degree is better, and it can reduce 15 wrong rules. The results of association rules generation show that teaching design should pay attention to the construction of network resources, and the reform of teaching content needs the promotion of high-quality teachers. The academic early warning model uses the GA-BP model to predict grades, and then formulates an early warning index based on the grades. The results show that the average accuracy rate of the prediction model is 89.12 %, which is better than other models, and the prediction accuracy rate of the potential early warning student group is >76.1 %. Compared with the final grades, the fitting degree of the prediction experiment results reaches 97.3 %, which shows that the performance of the model meets the needs of academic early warning.
{"title":"Optimization and implementation of management technology integrated with data analysis for college students' course evaluation and academic early warning","authors":"Xinxin Yang","doi":"10.1016/j.sasc.2025.200255","DOIUrl":"10.1016/j.sasc.2025.200255","url":null,"abstract":"<div><div>Educational managers need curriculum evaluation results and academic warnings to enrich educational management content. The study integrates data analysis technology into education and teaching, and uses association rules to mine the internal relationship of each dimension element of course evaluation. The results show that the performance of the Apriori algorithm with interest degree is better, and it can reduce 15 wrong rules. The results of association rules generation show that teaching design should pay attention to the construction of network resources, and the reform of teaching content needs the promotion of high-quality teachers. The academic early warning model uses the GA-BP model to predict grades, and then formulates an early warning index based on the grades. The results show that the average accuracy rate of the prediction model is 89.12 %, which is better than other models, and the prediction accuracy rate of the potential early warning student group is >76.1 %. Compared with the final grades, the fitting degree of the prediction experiment results reaches 97.3 %, which shows that the performance of the model meets the needs of academic early warning.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200255"},"PeriodicalIF":0.0,"publicationDate":"2025-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143929344","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-30DOI: 10.1016/j.sasc.2025.200272
Gong Yan
To improve the accuracy of basketball motion analysis, this study proposes a basketball motion recognition and tracking method based on an improved convolutional neural network. This method combines an intelligent sensor system with an improved dual-mode convolutional neural network to identify basketball motion steps; A tracking method based on the Northeast sky coordinate system was proposed to depict the motion trajectory of basketball players. The experimental results show that the average recognition accuracy of the improved convolutional neural network model is 99.3 %, which is superior to K-nearest neighbors and other models. This model structure can better capture the complexity and diversity of basketball footwork, improve recognition accuracy, and enhance generalization ability, while still maintaining high recognition accuracy in the face of new movements. The average error of linear trajectory tracking is 4.3 %, while the average errors of curved trajectory tracking in the X, Y, and Z directions are 4.1 %, 5.9 %, and 6.1 %, respectively. Research has shown that this method provides an effective approach for basketball analysis and training, which helps to improve the competitive level of basketball players.
{"title":"Basketball motion recognition and tracking method based on improved convolutional neural network","authors":"Gong Yan","doi":"10.1016/j.sasc.2025.200272","DOIUrl":"10.1016/j.sasc.2025.200272","url":null,"abstract":"<div><div>To improve the accuracy of basketball motion analysis, this study proposes a basketball motion recognition and tracking method based on an improved convolutional neural network. This method combines an intelligent sensor system with an improved dual-mode convolutional neural network to identify basketball motion steps; A tracking method based on the Northeast sky coordinate system was proposed to depict the motion trajectory of basketball players. The experimental results show that the average recognition accuracy of the improved convolutional neural network model is 99.3 %, which is superior to K-nearest neighbors and other models. This model structure can better capture the complexity and diversity of basketball footwork, improve recognition accuracy, and enhance generalization ability, while still maintaining high recognition accuracy in the face of new movements. The average error of linear trajectory tracking is 4.3 %, while the average errors of curved trajectory tracking in the X, Y, and Z directions are 4.1 %, 5.9 %, and 6.1 %, respectively. Research has shown that this method provides an effective approach for basketball analysis and training, which helps to improve the competitive level of basketball players.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200272"},"PeriodicalIF":0.0,"publicationDate":"2025-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143929348","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-30DOI: 10.1016/j.sasc.2025.200278
Minji Yin , Huixue Qu , Kailing Zhang
The dynamic visualization design of visual symbols is a complex task that requires finding a solution that meets user needs and has good visual effects. However, considering both the personalized needs of users and the dynamic visualization effect of design elements is a huge challenge. To address this challenge, this study proposes a dynamic visual symbol design method based on interactive genetic algorithm, which integrates real-time user participation mechanism and dynamic interactive feedback system to achieve intelligent and personalized design process. Compared with traditional genetic algorithms, this method introduces a dynamic fitness evaluation mechanism with direct user participation on the basis of traditional genetic algorithms. Users can evaluate the visual effects of individual populations in real time through a graphical interactive interface, and support real-time adjustment of parameters such as symbol form and motion trajectory. At the same time, it combines Bayesian probability models and Gaussian process proxy models to achieve dynamic capture and prediction of user preferences. The results show that the proposed method has a prediction accuracy of 89.1% and user satisfaction of 97.4% in dynamic visual symbol design, which is significantly improved compared to traditional genetic algorithms. This study provides new ideas for solving the dynamic design problem of multi-objective and high-dimensional visual symbols, which will help promote the development of the field of visual communication.
{"title":"Dynamic visualization design of visual symbols using interactive genetic algorithm","authors":"Minji Yin , Huixue Qu , Kailing Zhang","doi":"10.1016/j.sasc.2025.200278","DOIUrl":"10.1016/j.sasc.2025.200278","url":null,"abstract":"<div><div>The dynamic visualization design of visual symbols is a complex task that requires finding a solution that meets user needs and has good visual effects. However, considering both the personalized needs of users and the dynamic visualization effect of design elements is a huge challenge. To address this challenge, this study proposes a dynamic visual symbol design method based on interactive genetic algorithm, which integrates real-time user participation mechanism and dynamic interactive feedback system to achieve intelligent and personalized design process. Compared with traditional genetic algorithms, this method introduces a dynamic fitness evaluation mechanism with direct user participation on the basis of traditional genetic algorithms. Users can evaluate the visual effects of individual populations in real time through a graphical interactive interface, and support real-time adjustment of parameters such as symbol form and motion trajectory. At the same time, it combines Bayesian probability models and Gaussian process proxy models to achieve dynamic capture and prediction of user preferences. The results show that the proposed method has a prediction accuracy of 89.1% and user satisfaction of 97.4% in dynamic visual symbol design, which is significantly improved compared to traditional genetic algorithms. This study provides new ideas for solving the dynamic design problem of multi-objective and high-dimensional visual symbols, which will help promote the development of the field of visual communication.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200278"},"PeriodicalIF":0.0,"publicationDate":"2025-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144069346","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-30DOI: 10.1016/j.sasc.2025.200274
Ji Lu , Minjun Wu
Currently, music recommendation systems have significant limitations in user behavior analysis, resulting in lower accuracy in recommendations. To address these issues, we propose a music recommendation system based on user behavior and feature recognition, leveraging deep learning for training user data. User behavior sequences are inputted into an encoder to obtain datasets, detecting user preferences based on weight values. User gradients are derived through weighted partitioning, extracting user behavior intentions. User interest is statistically assessed based on the time spent listening to music, calculating a personalized music information matrix. Subset relevance is compared to achieve music information recommendations. Experimental comparisons with traditional systems show that the root mean square error (RMSE), mean square error (MSE), and mean absolute error (MAE) fluctuate between 0 and 1, with recommendation accuracy exceeding 87.5 % and peaking at 99 %, indicating excellent recommendation performance.
{"title":"Design and application of a music recommendation system based on user behavior and feature recognition","authors":"Ji Lu , Minjun Wu","doi":"10.1016/j.sasc.2025.200274","DOIUrl":"10.1016/j.sasc.2025.200274","url":null,"abstract":"<div><div>Currently, music recommendation systems have significant limitations in user behavior analysis, resulting in lower accuracy in recommendations. To address these issues, we propose a music recommendation system based on user behavior and feature recognition, leveraging deep learning for training user data. User behavior sequences are inputted into an encoder to obtain datasets, detecting user preferences based on weight values. User gradients are derived through weighted partitioning, extracting user behavior intentions. User interest is statistically assessed based on the time spent listening to music, calculating a personalized music information matrix. Subset relevance is compared to achieve music information recommendations. Experimental comparisons with traditional systems show that the root mean square error (RMSE), mean square error (MSE), and mean absolute error (MAE) fluctuate between 0 and 1, with recommendation accuracy exceeding 87.5 % and peaking at 99 %, indicating excellent recommendation performance.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200274"},"PeriodicalIF":0.0,"publicationDate":"2025-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143929347","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-30DOI: 10.1016/j.sasc.2025.200270
Sumei Zeng , Jinding Bao
With the rapid development of digital economy, the social process of "Internet+" has accelerated. The digital transformation of enterprise clusters has now become one of the key strategy to enhance the competitiveness and innovation ability of enterprises. A comprehensive and accurate analysis of the impact of digital transformation on the innovation performance of enterprises has a crucial impact on improving the economic benefits of enterprises. However, at present, the analysis of the innovation performance of digital transformation is not sufficient, and there are some limitations and one-sided points. Therefore, the research puts forward the impact model of the digital transformation of enterprise cluster based on cross-level analysis. The influence of the mechanism of enterprise cluster cooperation atmosphere on enterprise innovation performance is analyzed. The results of regression analysis show that the positive impact factor of enterprise cluster cooperation atmosphere on enterprise efficiency is enterprises (P < 0.05). To summarize the findings, both the collaborative environment and the identification of enterprise clusters have a significant positive impact on enterprise efficiency (P < 0.01). When the cooperation atmosphere within the enterprise cluster is good and the enterprise cluster members have a strong sense of identification to the cluster, the innovation performance and efficiency of the enterprise will be significantly improved. The research results reveal the influence of enterprise clusters on innovation performance, and provide a reference theory for the digital transformation of enterprise clusters in other regions.
{"title":"Analysis of the effects of digital transformation of enterprise clusters on innovation performance in the context of “Internet+”","authors":"Sumei Zeng , Jinding Bao","doi":"10.1016/j.sasc.2025.200270","DOIUrl":"10.1016/j.sasc.2025.200270","url":null,"abstract":"<div><div>With the rapid development of digital economy, the social process of \"Internet+\" has accelerated. The digital transformation of enterprise clusters has now become one of the key strategy to enhance the competitiveness and innovation ability of enterprises. A comprehensive and accurate analysis of the impact of digital transformation on the innovation performance of enterprises has a crucial impact on improving the economic benefits of enterprises. However, at present, the analysis of the innovation performance of digital transformation is not sufficient, and there are some limitations and one-sided points. Therefore, the research puts forward the impact model of the digital transformation of enterprise cluster based on cross-level analysis. The influence of the mechanism of enterprise cluster cooperation atmosphere on enterprise innovation performance is analyzed. The results of regression analysis show that the positive impact factor of enterprise cluster cooperation atmosphere on enterprise efficiency is enterprises (<em>P</em> < 0.05). To summarize the findings, both the collaborative environment and the identification of enterprise clusters have a significant positive impact on enterprise efficiency (<em>P</em> < 0.01). When the cooperation atmosphere within the enterprise cluster is good and the enterprise cluster members have a strong sense of identification to the cluster, the innovation performance and efficiency of the enterprise will be significantly improved. The research results reveal the influence of enterprise clusters on innovation performance, and provide a reference theory for the digital transformation of enterprise clusters in other regions.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200270"},"PeriodicalIF":0.0,"publicationDate":"2025-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143922863","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This study presents modeling and solving an ambulance routing problem to reduce the time spent providing patient services. A neural network based on a learning machine is used to train the data, and all the input data is analyzed and entered into the genetic algorithm to optimize the mathematical model. The input data was limited to a dataset that only included longitude, latitude, response time, and transportation cost. It was observed that most of the emergency requests were completed within 10 to 20 min. By decreasing the ambulance response time, more emergency cases can be dealt with in less time, preventing casualties and reducing costs. Also, the results showed that the average optimal response time for heart attack is 10 min; for car accidents, 23 min; for fire cases, 13 min; for fall cases, 25 min; for overdose cases, 29 min; and for sudden nervous attack cases, 30 min.
{"title":"Designing an ambulance routing optimization model using the combination of machine learning and genetic algorithm in conditions of uncertainty","authors":"Hamed Nozari , Agnieszka Szmelter-Jarosz , Hamid Reza Irani","doi":"10.1016/j.sasc.2025.200276","DOIUrl":"10.1016/j.sasc.2025.200276","url":null,"abstract":"<div><div>This study presents modeling and solving an ambulance routing problem to reduce the time spent providing patient services. A neural network based on a learning machine is used to train the data, and all the input data is analyzed and entered into the genetic algorithm to optimize the mathematical model. The input data was limited to a dataset that only included longitude, latitude, response time, and transportation cost. It was observed that most of the emergency requests were completed within 10 to 20 min. By decreasing the ambulance response time, more emergency cases can be dealt with in less time, preventing casualties and reducing costs. Also, the results showed that the average optimal response time for heart attack is 10 min; for car accidents, 23 min; for fire cases, 13 min; for fall cases, 25 min; for overdose cases, 29 min; and for sudden nervous attack cases, 30 min.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200276"},"PeriodicalIF":0.0,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143895587","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-29DOI: 10.1016/j.sasc.2025.200269
Yu Chen , Wenxing You , Lu Ou , Hui Tang
Urban resilience evaluates systems’ capacities to prepare for, adapt to, absorb, and recover from disruptions. Evaluation frameworks incorporate metrics like recovery speed, adaptive ability, and absorptive capacity. Assessing critical infrastructure interdependencies is challenging yet vital to limit failure propagation. While static assessments, multi-layer frameworks, and software like Hazus are used, limitations persist. Machine learning often focuses on infrastructure data for recovery monitoring. A common workflow entails acquiring and organizing data, then applying supervised, unsupervised, or reinforcement learning models. Supervised learning uses labeled data while unsupervised learning detects patterns in unlabeled data. Reinforcement learning optimizes rewards through trial-and-error interactions. Machine learning assists in meeting intensifying urbanization and climate change challenges. Leveraging advances in sensors, IoT, and computing enables tasks like image labeling and semantic segmentation. The techniques facilitate resilience through real-time data analytics for informed decision-making and responsive disaster management.
{"title":"A review of machine learning techniques for urban resilience research: The application and progress of different machine learning techniques in assessing and enhancing urban resilience","authors":"Yu Chen , Wenxing You , Lu Ou , Hui Tang","doi":"10.1016/j.sasc.2025.200269","DOIUrl":"10.1016/j.sasc.2025.200269","url":null,"abstract":"<div><div>Urban resilience evaluates systems’ capacities to prepare for, adapt to, absorb, and recover from disruptions. Evaluation frameworks incorporate metrics like recovery speed, adaptive ability, and absorptive capacity. Assessing critical infrastructure interdependencies is challenging yet vital to limit failure propagation. While static assessments, multi-layer frameworks, and software like Hazus are used, limitations persist. Machine learning often focuses on infrastructure data for recovery monitoring. A common workflow entails acquiring and organizing data, then applying supervised, unsupervised, or reinforcement learning models. Supervised learning uses labeled data while unsupervised learning detects patterns in unlabeled data. Reinforcement learning optimizes rewards through trial-and-error interactions. Machine learning assists in meeting intensifying urbanization and climate change challenges. Leveraging advances in sensors, IoT, and computing enables tasks like image labeling and semantic segmentation. The techniques facilitate resilience through real-time data analytics for informed decision-making and responsive disaster management.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200269"},"PeriodicalIF":0.0,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144105469","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-29DOI: 10.1016/j.sasc.2025.200277
Haijing Pan , Adzrool Idzwan bin Ismail , Asmidah Alwi , Massudi Mahmuddin
Introduction
With the rapid advancement of deep learning technologies, style transfer networks have demonstrated significant potential in the fields of image processing and creative design. Particularly in the realm of fashion design, style transfer techniques offer designers innovative tools to automatically generate diverse style designs, thereby enhancing creativity and diversity. However, existing style transfer methods still face challenges in balancing content preservation and style representation, as well as in computational efficiency. This study aims to explore a Neural Style Transfer (NST)-based model for fashion style transfer to address these issues and improve the efficiency and quality of fashion design.
Methodology
The proposed network architecture consists of three convolutional layers and one deconvolutional layer, designed to efficiently extract and integrate spatial features of fashion elements. Subsequently, the Visual Geometry Group (VGG)-Garment network architecture is employed for feature extraction and style fusion, with optimization algorithms generating high-quality fashion design images. Additionally, by introducing four semantic loss functions—content loss, style loss, color loss, and contour loss—the model ensures the preservation of the original design content while flexibly incorporating other visual styles.
Results
The experimental results demonstrate the following: (1) The proposed model excels in both style transfer effectiveness and computational efficiency. The style retention rate ranges from 82.11 % to 88.54 %. The content retention rate falls between 87.90 % and 92.56 %. These results indicate that the model effectively integrates diverse style elements while preserving the original design. (2) The proposed method outperforms three other models in terms of Peak Signal-to-Noise Ratio (PSNR) across all six fashion styles. Notably, in the "luxury" style, the PSNR value of the proposed method reaches 32.01, significantly higher than that of other models. (3) In terms of computational efficiency, the model generates a style-transferred fashion design image in an average of 15.23 s. The storage footprint is 251.45 MB, and the computational resource utilization rate is 60.78 %. These results show a significant improvement over traditional method.
Discussion
This study makes a significant contribution by proposing a model that enhances visual effects and design diversity. Additionally, it outperforms traditional methods in computational efficiency and resource utilization. This model provides a novel technical approach for the fashion design industry, effectively reducing design costs and enhancing the overall efficiency of the design process.
{"title":"The application and optimization of style transfer neural network based on deep learning in fashion design","authors":"Haijing Pan , Adzrool Idzwan bin Ismail , Asmidah Alwi , Massudi Mahmuddin","doi":"10.1016/j.sasc.2025.200277","DOIUrl":"10.1016/j.sasc.2025.200277","url":null,"abstract":"<div><h3>Introduction</h3><div>With the rapid advancement of deep learning technologies, style transfer networks have demonstrated significant potential in the fields of image processing and creative design. Particularly in the realm of fashion design, style transfer techniques offer designers innovative tools to automatically generate diverse style designs, thereby enhancing creativity and diversity. However, existing style transfer methods still face challenges in balancing content preservation and style representation, as well as in computational efficiency. This study aims to explore a Neural Style Transfer (NST)-based model for fashion style transfer to address these issues and improve the efficiency and quality of fashion design.</div></div><div><h3>Methodology</h3><div>The proposed network architecture consists of three convolutional layers and one deconvolutional layer, designed to efficiently extract and integrate spatial features of fashion elements. Subsequently, the Visual Geometry Group (VGG)-Garment network architecture is employed for feature extraction and style fusion, with optimization algorithms generating high-quality fashion design images. Additionally, by introducing four semantic loss functions—content loss, style loss, color loss, and contour loss—the model ensures the preservation of the original design content while flexibly incorporating other visual styles.</div></div><div><h3>Results</h3><div>The experimental results demonstrate the following: (1) The proposed model excels in both style transfer effectiveness and computational efficiency. The style retention rate ranges from 82.11 % to 88.54 %. The content retention rate falls between 87.90 % and 92.56 %. These results indicate that the model effectively integrates diverse style elements while preserving the original design. (2) The proposed method outperforms three other models in terms of Peak Signal-to-Noise Ratio (PSNR) across all six fashion styles. Notably, in the \"luxury\" style, the PSNR value of the proposed method reaches 32.01, significantly higher than that of other models. (3) In terms of computational efficiency, the model generates a style-transferred fashion design image in an average of 15.23 s. The storage footprint is 251.45 MB, and the computational resource utilization rate is 60.78 %. These results show a significant improvement over traditional method.</div></div><div><h3>Discussion</h3><div>This study makes a significant contribution by proposing a model that enhances visual effects and design diversity. Additionally, it outperforms traditional methods in computational efficiency and resource utilization. This model provides a novel technical approach for the fashion design industry, effectively reducing design costs and enhancing the overall efficiency of the design process.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200277"},"PeriodicalIF":0.0,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143898817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-28DOI: 10.1016/j.sasc.2025.200256
Jingui He , Hansi Ya
With the continuous growth of e-government data, traditional audit methods face increasing limitations in handling large-scale data, leading to low processing efficiency and insufficient accuracy. To address these challenges, this paper proposes a big data-driven evaluation and prediction model for e-government audit information. The proposed method is built on a Hadoop-based distributed computing platform, which supports heterogeneous data integration and efficient parallel processing. Furthermore, a parallel PSO-RF algorithm combining Particle Swarm Optimization (PSO) and Random Forest (RF) is designed to enhance classification performance and computational efficiency. Experiments are conducted using e-government audit data from a Chinese province collected between 2018 and 2020, covering 15 audit categories. The model performance is comprehensively evaluated using accuracy, recall, F1-score, and AUC metrics. Results demonstrate that the proposed parallel PSO-RF algorithm outperforms conventional RF and Support Vector Machine (SVM) approaches across multiple indicators, with a maximum prediction deviation of only 0.28 % compared to actual audit issue probabilities. This study not only improves the accuracy and efficiency of audit information processing but also provides a scalable technical approach and theoretical foundation for intelligent audit evaluation and risk assessment in e-government systems.
{"title":"Evaluation method of e-government audit information based on big data analysis","authors":"Jingui He , Hansi Ya","doi":"10.1016/j.sasc.2025.200256","DOIUrl":"10.1016/j.sasc.2025.200256","url":null,"abstract":"<div><div>With the continuous growth of e-government data, traditional audit methods face increasing limitations in handling large-scale data, leading to low processing efficiency and insufficient accuracy. To address these challenges, this paper proposes a big data-driven evaluation and prediction model for e-government audit information. The proposed method is built on a Hadoop-based distributed computing platform, which supports heterogeneous data integration and efficient parallel processing. Furthermore, a parallel PSO-RF algorithm combining Particle Swarm Optimization (PSO) and Random Forest (RF) is designed to enhance classification performance and computational efficiency. Experiments are conducted using e-government audit data from a Chinese province collected between 2018 and 2020, covering 15 audit categories. The model performance is comprehensively evaluated using accuracy, recall, F1-score, and AUC metrics. Results demonstrate that the proposed parallel PSO-RF algorithm outperforms conventional RF and Support Vector Machine (SVM) approaches across multiple indicators, with a maximum prediction deviation of only 0.28 % compared to actual audit issue probabilities. This study not only improves the accuracy and efficiency of audit information processing but also provides a scalable technical approach and theoretical foundation for intelligent audit evaluation and risk assessment in e-government systems.</div></div>","PeriodicalId":101205,"journal":{"name":"Systems and Soft Computing","volume":"7 ","pages":"Article 200256"},"PeriodicalIF":0.0,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143922862","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号