Xing Xu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Xing Xu has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Xuewen Xia
College of Physics and Information Engineering, Minnan Normal University, Zhangzhou 363000, China
5 shared publications - Yinglong Zhang
School of Physics and Information Engineering, Minnan Normal University, Zhangzhou, China
4 shared publications - Dongming Wang
College of Physics and Information Engineering, Minnan Normal University, Zhangzhou 363000, China
2 shared publications Xuewen Xia
School of Physics and Information Engineering, Minnan Normal University, Zhangzhou, China
2 shared publicationsYinglong Zhang
School of Physics and Information Engineering, Minnan Normal University, Zhangzhou, China
2 shared publications- Chengda Xu
Minnan Normal University, No. 36 Xianqian Road, Zhangzhou Fujian, 363000, China
1 shared publication
Latest Publications
- Open Access
- 13 June 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
FusionDiff: a dual-path diffusion-based framework for few-shot authenticity analysis of ceramic microstructures
The authenticity of ceramic components is closely tied to their microscopic structures, making automatic and accurate identification essential for quality control. However, this task is often constrained by the scarcity of labeled samples. This study investigates the potential of large-scale pretrained diffusion models as feature extractors, leveraging the rich visual priors embedded in their generative processes to provide a robust semantic foundation for small-sample learning. To address the limitations of the original U-Net in global representation modeling and the weak local-detail sensitivity of DeiT, we propose a dual-path fusion encoder, FusionDiff. Within a frozen Stable Diffusion V1.4 framework, CNN and adapter-enhanced DeiT paths operate in parallel and are deeply integrated via feature gating. Following a “self-supervised pretraining + supervised fine-tuning” paradigm, classification is performed using a Random Forest classifier. On our custom ceramic dataset, FusionDiff achieves a test accuracy of 99.07%, outperforming SD-CNN (97.44%), DeiT (96.30%), and ResNet50 (97.00%) under a unified self-supervised evaluation protocol. Even under extremely small-sample conditions ( $$n = 50$$ ), the model attains 90.7% validation accuracy, demonstrating competitive data efficiency and cross-domain generalization capability.- 13 June 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessFusionDiff: a dual-path diffusion-based framework for few-shot authenticity analysis of ceramic microstructures
The authenticity of ceramic components is closely tied to their microscopic structures, making automatic and accurate identification essential for quality control. However, this task is often constrained by the scarcity of labeled samples. This study investigates the potential of large-scale pretrained diffusion models as feature extractors, leveraging the rich visual priors embedded in their generative processes to provide a robust semantic foundation for small-sample learning. To address the limitations of the original U-Net in global representation modeling and the weak local-detail sensitivity of DeiT, we propose a dual-path fusion encoder, FusionDiff. Within a frozen Stable Diffusion V1.4 framework, CNN and adapter-enhanced DeiT paths operate in parallel and are deeply integrated via feature gating. Following a “self-supervised pretraining + supervised fine-tuning” paradigm, classification is performed using a Random Forest classifier. On our custom ceramic dataset, FusionDiff achieves a test accuracy of 99.07%, outperforming SD-CNN (97.44%), DeiT (96.30%), and ResNet50 (97.00%) under a unified self-supervised evaluation protocol. Even under extremely small-sample conditions ( $$n = 50$$ ), the model attains 90.7% validation accuracy, demonstrating competitive data efficiency and cross-domain generalization capability. - 10 February 2026
- Published by Elsevier in Neural Networks
- Vol. 199, 108711
Trainable-parameter-free structural-diversity message passing for graph neural networks
- 10 February 2026
- Published by Elsevier in Neural Networks
- Vol. 199, 108711
Trainable-parameter-free structural-diversity message passing for graph neural networks
- Open Access
- 1 January 2026
- Published by American Institute of Mathematical Sciences (AIMS) in Electronic Research Archive
- Vol. 34 (7) , 4889-4912
Edit-discrepancy-guided feature transformation in encoder-based GAN inversion for real image attribute editing
Image attribute editing based on generative adversarial networks (GANs) typically begins by mapping real images to the latent space of a pretrained StyleGAN, followed by manipulating the corresponding latent codes. However, low-rate latent codes suffer from an information bottleneck, making it challenging to faithfully reconstruct complex real images. Recent encoder-based methods enhance reconstruction by injecting high-rate features into intermediate generator layers to better preserve fine details, but they often yield misaligned details in the edited images. The primary reason is that these methods still rely on global linear transformations of high-rate features, which overlook the nonlinear and spatially localized nature of real edits. To this end, we build on a high-fidelity encoder-based GAN inversion backbone and introduce an additional adaptive feature editor that is specifically trained to convert high-rate features during editing so that fine details are correctly aligned with the edited image. The backbone refines the feature through a cross-attention mechanism and residual enhancement. Building on this, the feature editor employs a window-based cross-attention mechanism to extract a discrepancy signal between the original and edited generator features, which specifies both where to modify and what content to change. This signal is then fused into the feature through spatially adaptive modulation techniques, enabling region-selective attribute changes while preserving irrelevant details. Experiments on face and car benchmarks demonstrate that our method improves both reconstruction fidelity and editing quality compared to existing GAN inversion methods.
- 1 January 2026
- Published by American Institute of Mathematical Sciences (AIMS) in Electronic Research Archive
- Vol. 34 (7) , 4889-4912
Open AccessEdit-discrepancy-guided feature transformation in encoder-based GAN inversion for real image attribute editing
Image attribute editing based on generative adversarial networks (GANs) typically begins by mapping real images to the latent space of a pretrained StyleGAN, followed by manipulating the corresponding latent codes. However, low-rate latent codes suffer from an information bottleneck, making it challenging to faithfully reconstruct complex real images. Recent encoder-based methods enhance reconstruction by injecting high-rate features into intermediate generator layers to better preserve fine details, but they often yield misaligned details in the edited images. The primary reason is that these methods still rely on global linear transformations of high-rate features, which overlook the nonlinear and spatially localized nature of real edits. To this end, we build on a high-fidelity encoder-based GAN inversion backbone and introduce an additional adaptive feature editor that is specifically trained to convert high-rate features during editing so that fine details are correctly aligned with the edited image. The backbone refines the feature through a cross-attention mechanism and residual enhancement. Building on this, the feature editor employs a window-based cross-attention mechanism to extract a discrepancy signal between the original and edited generator features, which specifies both where to modify and what content to change. This signal is then fused into the feature through spatially adaptive modulation techniques, enabling region-selective attribute changes while preserving irrelevant details. Experiments on face and car benchmarks demonstrate that our method improves both reconstruction fidelity and editing quality compared to existing GAN inversion methods.
- Open Access
- 1 January 2026
- Published by IGI Global Scientific Publishing in International Journal of Cognitive Informatics and Natural Intelligence
- Vol. 20 (1) , 1-19
Research on Ceramic Recognition With Improved Dynamic Feature Fusion Module Based on YOLOv11
Accurate shape recognition of ceramic objects under resource constraints is vital for cultural-relic digitization and smart-museum construction. Conventional deep models, burdened by heavy parameters and slow inference, are hard to deploy on mobile or embedded devices. The authors present an ultra-lightweight yet highly efficient improved YOLOv11. MobileNetV4 serves as the backbone, Spatial Pyramid Pooling—Fast and positive-sensitive attention are integrated, a novel DFC3k2 module is innovatively designed and inserted to strengthen feature fusion, and spatial context convolution is combined to capture spatial context for multi-scale classification. On a public ceramic-shape dataset, the model reaches 97.67% top-1 accuracy, 1.69 percentage points (pp) higher than the YOLOv11-classification (cls) baseline (95.98%). Parameters rise only 2.08 meters (M), from 1.56 M to 3.64 M, still within the lightweight range. Mean inference time is 17.85 ms, while precision, recall, and F1 score remain excellent. Experiments show a clear accuracy gain with minimal parametric cost, balancing lightness and efficiency and offering real-time relic recognition capability in resource-limited scenarios.
- 1 January 2026
- Published by IGI Global Scientific Publishing in International Journal of Cognitive Informatics and Natural Intelligence
- Vol. 20 (1) , 1-19
Open AccessResearch on Ceramic Recognition With Improved Dynamic Feature Fusion Module Based on YOLOv11
Accurate shape recognition of ceramic objects under resource constraints is vital for cultural-relic digitization and smart-museum construction. Conventional deep models, burdened by heavy parameters and slow inference, are hard to deploy on mobile or embedded devices. The authors present an ultra-lightweight yet highly efficient improved YOLOv11. MobileNetV4 serves as the backbone, Spatial Pyramid Pooling—Fast and positive-sensitive attention are integrated, a novel DFC3k2 module is innovatively designed and inserted to strengthen feature fusion, and spatial context convolution is combined to capture spatial context for multi-scale classification. On a public ceramic-shape dataset, the model reaches 97.67% top-1 accuracy, 1.69 percentage points (pp) higher than the YOLOv11-classification (cls) baseline (95.98%). Parameters rise only 2.08 meters (M), from 1.56 M to 3.64 M, still within the lightweight range. Mean inference time is 17.85 ms, while precision, recall, and F1 score remain excellent. Experiments show a clear accuracy gain with minimal parametric cost, balancing lightness and efficiency and offering real-time relic recognition capability in resource-limited scenarios.
- Open Access
- 11 September 2024
- Published by MDPI AG in Algorithms
- Vol. 17 (9) , 407
Interactive 3D Vase Design Based on Gradient Boosting Decision Trees
Traditionally, ceramic design began with sketches on rough paper and later evolved into using CAD software for more complex designs and simulations. With technological advancements, optimization algorithms have gradually been introduced into ceramic design to enhance design efficiency and creative diversity. The use of Interactive Genetic Algorithms (IGAs) for ceramic design is a new approach, but an IGA requires a significant amount of user evaluation, which can result in user fatigue. To overcome this problem, this paper introduces the LightGBM algorithm and the CatBoost algorithm to improve the IGA because they have excellent predictive capabilities that can assist users in evaluations. The algorithms are also applied to a vase design platform for validation. First, bicubic Bézier surfaces are used for modeling, and the genetic encoding of the vase is designed with appropriate evolutionary operators selected. Second, user data from the online platform are collected to train and optimize the LightGBM and CatBoost algorithms. Finally, LightGBM and CatBoost are combined with an IGA and applied to the vase design platform to verify their effectiveness. Comparing the improved algorithm to traditional IGAs, KD trees, Random Forest, and XGBoost, it is found that IGAs improve with LightGBM, and CatBoost performs better overall, requiring fewer evaluations and less time. Its R2 is higher than other proxy models, achieving 0.816 and 0.839, respectively. The improved method proposed in this paper can effectively alleviate user fatigue and enhance the user experience in product design participation.
- 11 September 2024
- Published by MDPI AG in Algorithms
- Vol. 17 (9) , 407
Open AccessInteractive 3D Vase Design Based on Gradient Boosting Decision Trees
Traditionally, ceramic design began with sketches on rough paper and later evolved into using CAD software for more complex designs and simulations. With technological advancements, optimization algorithms have gradually been introduced into ceramic design to enhance design efficiency and creative diversity. The use of Interactive Genetic Algorithms (IGAs) for ceramic design is a new approach, but an IGA requires a significant amount of user evaluation, which can result in user fatigue. To overcome this problem, this paper introduces the LightGBM algorithm and the CatBoost algorithm to improve the IGA because they have excellent predictive capabilities that can assist users in evaluations. The algorithms are also applied to a vase design platform for validation. First, bicubic Bézier surfaces are used for modeling, and the genetic encoding of the vase is designed with appropriate evolutionary operators selected. Second, user data from the online platform are collected to train and optimize the LightGBM and CatBoost algorithms. Finally, LightGBM and CatBoost are combined with an IGA and applied to the vase design platform to verify their effectiveness. Comparing the improved algorithm to traditional IGAs, KD trees, Random Forest, and XGBoost, it is found that IGAs improve with LightGBM, and CatBoost performs better overall, requiring fewer evaluations and less time. Its R2 is higher than other proxy models, achieving 0.816 and 0.839, respectively. The improved method proposed in this paper can effectively alleviate user fatigue and enhance the user experience in product design participation.
- Open Access
- 21 June 2024
- Published by MDPI AG in Mathematics
- Vol. 12 (13) , 1932
3D Vase Design Based on Interactive Genetic Algorithm and Enhanced XGBoost Model
The human–computer interaction attribute of the interactive genetic algorithm (IGA) allows users to participate in the product design process for which the product needs to be evaluated, and requiring a large number of evaluations would lead to user fatigue. To address this issue, this paper utilizes an XGBoost proxy model modified by particle swarm optimization and the graphical interaction mechanism (GIM) to construct an improved interactive genetic algorithm (PXG-IGA), and then the PXG-IGA is applied to 3D vase design. Firstly, the 3D vase shape has been designed by using a bicubic Bézier surface, and the individual genetic code is binary and includes three parts: the vase control points, the vase height, and the texture picture. Secondly, the XGBoost evaluation of the proxy model has been constructed by collecting user online evaluation data, and the particle swarm optimization algorithm has been used to optimize the hyperparameters of XGBoost. Finally, the GIM has been introduced after several generations, allowing users to change product styles independently to better meet users’ expectations. Based on the PXG-IGA, an online 3D vase design platform has been developed and compared to the traditional IGA, KD tree, random forest, and standard XGBoost proxy models. Compared with the traditional IGA, the number of evaluations has been reduced by 58.3% and the evaluation time has been reduced by 46.4%. Compared with other proxy models, the accuracy of predictions has been improved up from 1.3% to 20.2%. To a certain extent, the PXG-IGA reduces users’ operation fatigue and provides new ideas for improving user experience and product design efficiency.
- 21 June 2024
- Published by MDPI AG in Mathematics
- Vol. 12 (13) , 1932
Open Access3D Vase Design Based on Interactive Genetic Algorithm and Enhanced XGBoost Model
The human–computer interaction attribute of the interactive genetic algorithm (IGA) allows users to participate in the product design process for which the product needs to be evaluated, and requiring a large number of evaluations would lead to user fatigue. To address this issue, this paper utilizes an XGBoost proxy model modified by particle swarm optimization and the graphical interaction mechanism (GIM) to construct an improved interactive genetic algorithm (PXG-IGA), and then the PXG-IGA is applied to 3D vase design. Firstly, the 3D vase shape has been designed by using a bicubic Bézier surface, and the individual genetic code is binary and includes three parts: the vase control points, the vase height, and the texture picture. Secondly, the XGBoost evaluation of the proxy model has been constructed by collecting user online evaluation data, and the particle swarm optimization algorithm has been used to optimize the hyperparameters of XGBoost. Finally, the GIM has been introduced after several generations, allowing users to change product styles independently to better meet users’ expectations. Based on the PXG-IGA, an online 3D vase design platform has been developed and compared to the traditional IGA, KD tree, random forest, and standard XGBoost proxy models. Compared with the traditional IGA, the number of evaluations has been reduced by 58.3% and the evaluation time has been reduced by 46.4%. Compared with other proxy models, the accuracy of predictions has been improved up from 1.3% to 20.2%. To a certain extent, the PXG-IGA reduces users’ operation fatigue and provides new ideas for improving user experience and product design efficiency.
- Open Access
- 8 July 2022
- Published by Wiley in Computational Intelligence and Neuroscience
- Vol. 2022, 1-11
Improved Interactive Genetic Algorithm for Three-Dimensional Vase Modeling Design
Interactive genetic algorithm (IGA) is an effective way to help users with product design optimization. However, in this process, users need to evaluate the fitness of all individuals in each generation. It will cause users’ fatigue when users cannot find satisfactory products after multi-generation evaluations. To solve this problem, an improved interactive genetic algorithm (IGA-KDTGIM) is proposed, which combines K-dimensional tree surrogate model and a graphic interaction mechanism. In this algorithm, the K-dimensional tree surrogate model is built on the basis of users' historical evaluation information to assist the user's evaluation, so as to reduce the times of users' evaluation. At the same time, users are allowed to interact with the graphic interface to adjust the shape of the individual, so as to increase users' creation fun and to make the evolution direction of the population conform to users' expectations. The IGA-KDTGIM is applied to the 3D vase design system and independently experimented with IGA, IGA-KDT, and IGA-GIM, respectively. The average fitness, maximum average fitness, and evaluation times of statistical data were compared and analyzed. Compared with traditional IGA, the number of evaluations required by users decreased by 60.0%, and the average fitness of the population increased by 15.0%. The results show that this method can reduce the users' operation fatigue and improve the ability of finding satisfactory solutions to a certain extent.
- 8 July 2022
- Published by Wiley in Computational Intelligence and Neuroscience
- Vol. 2022, 1-11
Open AccessImproved Interactive Genetic Algorithm for Three-Dimensional Vase Modeling Design
Interactive genetic algorithm (IGA) is an effective way to help users with product design optimization. However, in this process, users need to evaluate the fitness of all individuals in each generation. It will cause users’ fatigue when users cannot find satisfactory products after multi-generation evaluations. To solve this problem, an improved interactive genetic algorithm (IGA-KDTGIM) is proposed, which combines K-dimensional tree surrogate model and a graphic interaction mechanism. In this algorithm, the K-dimensional tree surrogate model is built on the basis of users' historical evaluation information to assist the user's evaluation, so as to reduce the times of users' evaluation. At the same time, users are allowed to interact with the graphic interface to adjust the shape of the individual, so as to increase users' creation fun and to make the evolution direction of the population conform to users' expectations. The IGA-KDTGIM is applied to the 3D vase design system and independently experimented with IGA, IGA-KDT, and IGA-GIM, respectively. The average fitness, maximum average fitness, and evaluation times of statistical data were compared and analyzed. Compared with traditional IGA, the number of evaluations required by users decreased by 60.0%, and the average fitness of the population increased by 15.0%. The results show that this method can reduce the users' operation fatigue and improve the ability of finding satisfactory solutions to a certain extent.
- Open Access
- 1 January 2022
- Published by IGI Global Scientific Publishing in International Journal of Cognitive Informatics and Natural Intelligence
- Vol. 16 (1) , 1-22
The Construction and Optimization of an AI Education Evaluation Indicator Based on Intelligent Algorithms
The basic tool in the analytic hierarchy process (AHP) is the complete judgment matrix. To address the weakness of the AHP in determining weight in the comprehensive evaluation system, the particle swarm optimization (PSO)-AHP model proposed in this paper is based on the PSO in the meta-heuristic algorithm. The model was used to solve the indicator weights in the evaluation system of AI education in primary and secondary schools in Fujian Province and was compared with the genetic algorithm and war strategy optimization algorithm. From the comparison results, the PSO-AHP optimization is more effective among the three algorithms, and the indicator consistency can be improved by about 30%. They are both effective in solving the problem that once the judgment matrix is given in the AHP, the weights and indicator consistency cannot be improved. Finally, the results were tested by Friedman statistics to prove the viability of the proposed algorithm.
- 1 January 2022
- Published by IGI Global Scientific Publishing in International Journal of Cognitive Informatics and Natural Intelligence
- Vol. 16 (1) , 1-22
Open AccessThe Construction and Optimization of an AI Education Evaluation Indicator Based on Intelligent Algorithms
The basic tool in the analytic hierarchy process (AHP) is the complete judgment matrix. To address the weakness of the AHP in determining weight in the comprehensive evaluation system, the particle swarm optimization (PSO)-AHP model proposed in this paper is based on the PSO in the meta-heuristic algorithm. The model was used to solve the indicator weights in the evaluation system of AI education in primary and secondary schools in Fujian Province and was compared with the genetic algorithm and war strategy optimization algorithm. From the comparison results, the PSO-AHP optimization is more effective among the three algorithms, and the indicator consistency can be improved by about 30%. They are both effective in solving the problem that once the judgment matrix is given in the AHP, the weights and indicator consistency cannot be improved. Finally, the results were tested by Friedman statistics to prove the viability of the proposed algorithm.
- 6 February 2014
- Published by Trans Tech Publications, Ltd. in Applied Mechanics and Materials
- Vol. 513-517, 391-394
Cloud Task and Virtual Machine Allocation Strategy Based on Simulated Annealing-Genetic Algorithm
In order to solve and optimize the task and virtual machine allocation strategy in cloud computing environment, firstly allocation strategy mathematical model, the target of which is the total running time, is established, then a simulated annealing and genetic hybrid algorithm is proposed to solve the mathematical model. The integer coding, a crossover operator, two kinds of mutation operators and the selection mechanism based on simulated annealing strategy are applied in the hybrid algorithm. In the experiments, three sets of data are used to verify the performance of the hybrid algorithm in the Cloudsim software. And the experimental results show that the series of cloud tasks can effectively be assigned to the virtual machine by the hybrid algorithm and the total running time is also minimized by the algorithm.
- 6 February 2014
- Published by Trans Tech Publications, Ltd. in Applied Mechanics and Materials
- Vol. 513-517, 391-394
Cloud Task and Virtual Machine Allocation Strategy Based on Simulated Annealing-Genetic Algorithm
In order to solve and optimize the task and virtual machine allocation strategy in cloud computing environment, firstly allocation strategy mathematical model, the target of which is the total running time, is established, then a simulated annealing and genetic hybrid algorithm is proposed to solve the mathematical model. The integer coding, a crossover operator, two kinds of mutation operators and the selection mechanism based on simulated annealing strategy are applied in the hybrid algorithm. In the experiments, three sets of data are used to verify the performance of the hybrid algorithm in the Cloudsim software. And the experimental results show that the series of cloud tasks can effectively be assigned to the virtual machine by the hybrid algorithm and the total running time is also minimized by the algorithm.