Applied Soft Computing最新文献_第2页

Domain Adaptation via Feature Disentanglement for cross-domain image classification

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2025-02-17 DOI: 10.1016/j.asoc.2025.112868

Zhi-Ze Wu , Chang-Jiang Du , Xin-Qi Wang , Le Zou , Fan Cheng , Teng Li , Fu-Dong Nian , Thomas Weise , Xiao-Feng Wang

Image classification is an important application area of soft computing. In many real-world application scenarios, image classifiers are applied to domains that differ from the original training data. This so-called domain shift significantly reduces classification accuracy. To tackle this issue, unsupervised domain adaptation (UDA) techniques have been developed to bridge the gap between source and target domains. These techniques achieve this by transferring knowledge from a labeled source domain to an unlabeled target domain. We develop a novel and effective coarse-to-fine domain adaptation method called Domain Adaptation via Feature Disentanglement (DAFD), which has two new key components: First, our Class-Relevant Feature Selection (CRFS) module disentangles class-relevant features from class-irrelevant ones. This prevents the network from overfitting to irrelevant data and enhances its focus on crucial information for accurate classification. This reduces the complexity of domain alignment, which improves the classification accuracy on the target domain. Second, our Dynamic Local Maximum Mean Discrepancy module DLMMD achieves a fine-grained feature alignment by minimizing the discrepancy among class-relevant features from different domains. The alignment process now becomes more adaptive and contextually sensitive, enhancing the ability of the model to recognize domain-specific patterns and characteristics. The combination of the CRFS and DLMMD modules results in an effective alignment of class-relevant features. Domain knowledge is successfully transferred from the source to the target domain. Our comprehensive experiments on four standard datasets demonstrate that DAFD is robust and highly effective in cross-domain image classification tasks.

图像分类是软计算的一个重要应用领域。在现实世界的许多应用场景中，图像分类器被应用于不同于原始训练数据的领域。这种所谓的领域偏移会大大降低分类精度。为解决这一问题，人们开发了无监督域适应（UDA）技术，以弥合源域和目标域之间的差距。这些技术通过将知识从有标注的源域转移到无标注的目标域来实现这一目标。我们开发了一种新颖而有效的从粗到细的域适应方法，称为 "通过特征分离进行域适应"（DAFD），它有两个新的关键组成部分：首先，我们的类相关特征选择（CRFS）模块将类相关特征与类不相关特征分离开来。这可以防止网络过度拟合不相关的数据，并使其更加专注于准确分类的关键信息。这降低了领域对齐的复杂性，从而提高了目标领域的分类准确性。其次，我们的动态局部最大均值差异模块 DLMMD 通过最小化来自不同领域的类相关特征之间的差异，实现了细粒度特征配准。现在，对齐过程变得更具适应性和语境敏感性，从而提高了模型识别特定领域模式和特征的能力。CRFS 模块和 DLMMD 模块的结合实现了类相关特征的有效配准。领域知识成功地从源领域转移到了目标领域。我们在四个标准数据集上进行的综合实验证明，DAFD 在跨领域图像分类任务中是稳健而高效的。

{"title":"Domain Adaptation via Feature Disentanglement for cross-domain image classification","authors":"Zhi-Ze Wu , Chang-Jiang Du , Xin-Qi Wang , Le Zou , Fan Cheng , Teng Li , Fu-Dong Nian , Thomas Weise , Xiao-Feng Wang","doi":"10.1016/j.asoc.2025.112868","DOIUrl":"10.1016/j.asoc.2025.112868","url":null,"abstract":"<div><div>Image classification is an important application area of soft computing. In many real-world application scenarios, image classifiers are applied to domains that differ from the original training data. This so-called domain shift significantly reduces classification accuracy. To tackle this issue, unsupervised domain adaptation (UDA) techniques have been developed to bridge the gap between source and target domains. These techniques achieve this by transferring knowledge from a labeled source domain to an unlabeled target domain. We develop a novel and effective coarse-to-fine domain adaptation method called Domain Adaptation via Feature Disentanglement (DAFD), which has two new key components: First, our Class-Relevant Feature Selection (CRFS) module disentangles class-relevant features from class-irrelevant ones. This prevents the network from overfitting to irrelevant data and enhances its focus on crucial information for accurate classification. This reduces the complexity of domain alignment, which improves the classification accuracy on the target domain. Second, our Dynamic Local Maximum Mean Discrepancy module DLMMD achieves a fine-grained feature alignment by minimizing the discrepancy among class-relevant features from different domains. The alignment process now becomes more adaptive and contextually sensitive, enhancing the ability of the model to recognize domain-specific patterns and characteristics. The combination of the CRFS and DLMMD modules results in an effective alignment of class-relevant features. Domain knowledge is successfully transferred from the source to the target domain. Our comprehensive experiments on four standard datasets demonstrate that DAFD is robust and highly effective in cross-domain image classification tasks.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112868"},"PeriodicalIF":7.2,"publicationDate":"2025-02-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143445320","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

On generalized Sugeno’s class generator and parametrized intuitionistic fuzzy approach for enhancing low-light images

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2025-02-17 DOI: 10.1016/j.asoc.2025.112865

Maheshkumar C.V. , David Raj M. , Saraswathi D.

Enhancing low-light images poses a significant challenge in terms of pixel distortion, color degradation, detail loss, over enhancement and noise amplification, particularly in images that have both low light and normal light region. In recent years, researchers have increasingly turned their attention to intuitionistic fuzzy set based approaches for low light image enhancement due to their flexibility in the representation of a pixel. In this work, the generalized Sugeno’s class of generating function is proposed. Since the parameter value in the existing generating functions lies in an unbounded interval, it is difficult to find the best parameter value. By using the proposed generalized version, a few intuitionistic generating functions are analyzed where the parameter value lies in a bounded interval. A searching algorithm is also proposed to find the parameter value that maximizes the entropy of an image for any membership and generating function. Regardless of the number of decimals, the proposed approach finds the best parameter value iteratively. Then, in HSI color space, an enhancement model is designed utilizing the intuitionistic fuzzy image achieved using best parameter value and contrast-limited adaptive histogram equalization. The proposed method performs better compared to the state-of-the-art models. Also, seven image quality mathematical metrics — entropy, SSIM, correlation coefficient

(r)

, PSNR, AMBE, number of edge pixels

(N_{g})

and the fitness function are implemented to compare the proposed and state-of-the-art models.

{"title":"On generalized Sugeno’s class generator and parametrized intuitionistic fuzzy approach for enhancing low-light images","authors":"Maheshkumar C.V. , David Raj M. , Saraswathi D.","doi":"10.1016/j.asoc.2025.112865","DOIUrl":"10.1016/j.asoc.2025.112865","url":null,"abstract":"<div><div>Enhancing low-light images poses a significant challenge in terms of pixel distortion, color degradation, detail loss, over enhancement and noise amplification, particularly in images that have both low light and normal light region. In recent years, researchers have increasingly turned their attention to intuitionistic fuzzy set based approaches for low light image enhancement due to their flexibility in the representation of a pixel. In this work, the generalized Sugeno’s class of generating function is proposed. Since the parameter value in the existing generating functions lies in an unbounded interval, it is difficult to find the best parameter value. By using the proposed generalized version, a few intuitionistic generating functions are analyzed where the parameter value lies in a bounded interval. A searching algorithm is also proposed to find the parameter value that maximizes the entropy of an image for any membership and generating function. Regardless of the number of decimals, the proposed approach finds the best parameter value iteratively. Then, in HSI color space, an enhancement model is designed utilizing the intuitionistic fuzzy image achieved using best parameter value and contrast-limited adaptive histogram equalization. The proposed method performs better compared to the state-of-the-art models. Also, seven image quality mathematical metrics — entropy, SSIM, correlation coefficient <span><math><mrow><mo>(</mo><mi>r</mi><mo>)</mo></mrow></math></span>, PSNR, AMBE, number of edge pixels <span><math><mrow><mo>(</mo><msub><mrow><mi>N</mi></mrow><mrow><mi>g</mi></mrow></msub><mo>)</mo></mrow></math></span> and the fitness function are implemented to compare the proposed and state-of-the-art models.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112865"},"PeriodicalIF":7.2,"publicationDate":"2025-02-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143471604","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Parallel attribute reduction in high-dimensional data: An efficient MapReduce strategy with fuzzy discernibility matrix

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2025-02-17 DOI: 10.1016/j.asoc.2025.112870

Pandu Sowkuntla , P.S.V.S. Sai Prasad

The hybrid paradigm of fuzzy-rough set theory, which combines fuzzy and rough sets, has proven effective in attribute reduction for hybrid decision systems encompassing both numerical and categorical attributes. However, current parallel/distributed approaches are limited to handling datasets with either categorical or numerical attributes and often rely on fuzzy dependency measures. There exists little research on parallel/distributed attribute reduction for large-scale hybrid decision systems. The challenge of handling high-dimensional data in hybrid decision systems necessitates efficient distributed computing techniques to ensure scalability and performance. MapReduce, a widely used framework for distributed processing, provides an organized approach to handling large-scale data. Despite its potential, there is a noticeable lack of attribute reduction techniques that leverage MapReduce’s capabilities with a fuzzy discernibility matrix, which can significantly improve the efficiency of processing high-dimensional hybrid datasets. This paper introduces a vertically partitioned fuzzy discernibility matrix within the MapReduce computation model to address the high dimensionality of hybrid datasets. The proposed MapReduce strategy for attribute reduction minimizes data movement during the shuffle and sort phase, overcoming limitations present in existing approaches. Furthermore, the method’s efficiency is enhanced by integrating a feature known as SAT-region removal, which removes matrix entries that satisfy the maximum satisfiability conditions during the attribute reduction process. Extensive experimental analysis validates the proposed method, demonstrating its superior performance compared to recent parallel/distributed methods in attribute reduction.

{"title":"Parallel attribute reduction in high-dimensional data: An efficient MapReduce strategy with fuzzy discernibility matrix","authors":"Pandu Sowkuntla , P.S.V.S. Sai Prasad","doi":"10.1016/j.asoc.2025.112870","DOIUrl":"10.1016/j.asoc.2025.112870","url":null,"abstract":"<div><div>The hybrid paradigm of fuzzy-rough set theory, which combines fuzzy and rough sets, has proven effective in attribute reduction for hybrid decision systems encompassing both numerical and categorical attributes. However, current parallel/distributed approaches are limited to handling datasets with either categorical or numerical attributes and often rely on fuzzy dependency measures. There exists little research on parallel/distributed attribute reduction for large-scale hybrid decision systems. The challenge of handling high-dimensional data in hybrid decision systems necessitates efficient distributed computing techniques to ensure scalability and performance. MapReduce, a widely used framework for distributed processing, provides an organized approach to handling large-scale data. Despite its potential, there is a noticeable lack of attribute reduction techniques that leverage MapReduce’s capabilities with a fuzzy discernibility matrix, which can significantly improve the efficiency of processing high-dimensional hybrid datasets. This paper introduces a vertically partitioned fuzzy discernibility matrix within the MapReduce computation model to address the high dimensionality of hybrid datasets. The proposed MapReduce strategy for attribute reduction minimizes data movement during the shuffle and sort phase, overcoming limitations present in existing approaches. Furthermore, the method’s efficiency is enhanced by integrating a feature known as SAT-region removal, which removes matrix entries that satisfy the maximum satisfiability conditions during the attribute reduction process. Extensive experimental analysis validates the proposed method, demonstrating its superior performance compared to recent parallel/distributed methods in attribute reduction.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112870"},"PeriodicalIF":7.2,"publicationDate":"2025-02-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143455106","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Data-driven evolutionary algorithms based on initialization selection strategies, POX crossover and multi-point random mutation for flexible job shop scheduling problems

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2025-02-17 DOI: 10.1016/j.asoc.2025.112901

Ruxin Zhao , Lixiang Fu , Jiajie Kang , Chang Liu , Wei Wang , Haizhou Wu , Yang Shi , Chao Jiang , Rui Wang

In the fields of manufacturing and production, the precise solution of the flexible job shop scheduling problem (FJSP) is crucial for improving production efficiency and optimizing resource allocation. However, the complexity of FJSP often leads traditional optimization methods to face high computational costs and lengthy processing times. To address this problem, we propose a data-driven evolutionary algorithm based on initialization selection strategies, POX crossover, and multi-point random mutation (DDEA-PMI). This algorithm replaces the real objective function by constructing a radial basis function (RBF) surrogate model to reduce expensive computational costs and shorten solution time. In the process of solving FJSP, we use global selection (GS), local selection (LS), and random selection (RS) initialization selection strategies to obtain an initial population with high diversity. In order to reduce the generation of infeasible solutions, we use the POX crossover operator, which selects partial gene sequences from the parent generation and maps them to the offspring to preserve excellent features and ensure the feasibility of the solution. In addition, we design a multi-point random mutation operation to enhance the diversity of the population. Through the multi-point mutation strategy, it is able to explore more comprehensively in the solution space to increase the possibility of finding the optimal solution. To verify the effectiveness of DDEA-PMI, we compare it with three same types of data-driven evolutionary algorithms. We compare and analyze the DDEA-PMI with three algorithms after removing one of our proposed strategies. The experimental results show that DDEA-PMI is effective and has advantages in solving FJSP.

{"title":"Data-driven evolutionary algorithms based on initialization selection strategies, POX crossover and multi-point random mutation for flexible job shop scheduling problems","authors":"Ruxin Zhao , Lixiang Fu , Jiajie Kang , Chang Liu , Wei Wang , Haizhou Wu , Yang Shi , Chao Jiang , Rui Wang","doi":"10.1016/j.asoc.2025.112901","DOIUrl":"10.1016/j.asoc.2025.112901","url":null,"abstract":"<div><div>In the fields of manufacturing and production, the precise solution of the flexible job shop scheduling problem (FJSP) is crucial for improving production efficiency and optimizing resource allocation. However, the complexity of FJSP often leads traditional optimization methods to face high computational costs and lengthy processing times. To address this problem, we propose a data-driven evolutionary algorithm based on initialization selection strategies, POX crossover, and multi-point random mutation (DDEA-PMI). This algorithm replaces the real objective function by constructing a radial basis function (RBF) surrogate model to reduce expensive computational costs and shorten solution time. In the process of solving FJSP, we use global selection (GS), local selection (LS), and random selection (RS) initialization selection strategies to obtain an initial population with high diversity. In order to reduce the generation of infeasible solutions, we use the POX crossover operator, which selects partial gene sequences from the parent generation and maps them to the offspring to preserve excellent features and ensure the feasibility of the solution. In addition, we design a multi-point random mutation operation to enhance the diversity of the population. Through the multi-point mutation strategy, it is able to explore more comprehensively in the solution space to increase the possibility of finding the optimal solution. To verify the effectiveness of DDEA-PMI, we compare it with three same types of data-driven evolutionary algorithms. We compare and analyze the DDEA-PMI with three algorithms after removing one of our proposed strategies. The experimental results show that DDEA-PMI is effective and has advantages in solving FJSP.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112901"},"PeriodicalIF":7.2,"publicationDate":"2025-02-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464774","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A directed batch growing self-organizing map based niching differential evolution for multimodal optimization problems

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2025-02-17 DOI: 10.1016/j.asoc.2025.112862

Mahesh Shankar , Palaniappan Ramu , Kalyanmoy Deb

Many real-world optimization problems naturally result in multiple optimal solutions, thereby falling in the class of multimodal optimization problems (MMOPs). A task of finding a plurality of optimal solutions for MMOPs comes under the scope of multimodal optimization algorithms (MMOAs). To solve MMOPs, niching techniques are usually employed by proactively modifying standard evolutionary algorithms (EAs) to form stable subpopulations around multiple niches within their evolving populations. This way, each optimum can germinate and eventually help form a cloud of solutions around each optimum parallely, thereby finding multiple (but a finite number of) optima simultaneously. However, several existing niching techniques suffer from common drawbacks, such as sensitivity with niching parameters or poor performance on high-dimensional problems. An efficient niching technique needs an effective population partitioning method around distinct leading solutions representing each optimum. In this paper, we propose a directed batch growing self-organizing map based niching differential evolution (DBGSOM-NDE). For this purpose, a standard differential evolution (DE) method is divided into two overlapping phases: (i) population-wide search (PS) and (ii) niche-wide search (NS). PS executes neighborhood search around each individual, promoting exploration, while NS explores only the leaders, thus reducing the effect of exploration for a better search intensification around the leaders using a Cauchy-distribution based local search to improve them. We evaluate the role of each operator of the proposed approach DBGSOM-NDE and compare its performance with a number of state-of-the-art niching techniques demonstrating its competitiveness and superiority, especially on high-dimensional and nonlinear problems taken from the existing literature. Finally, a hyper-parametric study is provided demonstrating weak dependence of them to the algorithm’s performance.

{"title":"A directed batch growing self-organizing map based niching differential evolution for multimodal optimization problems","authors":"Mahesh Shankar , Palaniappan Ramu , Kalyanmoy Deb","doi":"10.1016/j.asoc.2025.112862","DOIUrl":"10.1016/j.asoc.2025.112862","url":null,"abstract":"<div><div>Many real-world optimization problems naturally result in multiple optimal solutions, thereby falling in the class of multimodal optimization problems (MMOPs). A task of finding a plurality of optimal solutions for MMOPs comes under the scope of multimodal optimization algorithms (MMOAs). To solve MMOPs, <em>niching</em> techniques are usually employed by proactively modifying standard evolutionary algorithms (EAs) to form stable subpopulations around multiple niches within their evolving populations. This way, each optimum can germinate and eventually help form a cloud of solutions around each optimum parallely, thereby finding multiple (but a finite number of) optima simultaneously. However, several existing niching techniques suffer from common drawbacks, such as sensitivity with niching parameters or poor performance on high-dimensional problems. An efficient niching technique needs an effective population partitioning method around distinct leading solutions representing each optimum. In this paper, we propose a directed batch growing self-organizing map based niching differential evolution (DBGSOM-NDE). For this purpose, a standard differential evolution (DE) method is divided into two overlapping phases: (i) population-wide search (PS) and (ii) niche-wide search (NS). PS executes neighborhood search around each individual, promoting exploration, while NS explores only the leaders, thus reducing the effect of exploration for a better search intensification around the leaders using a Cauchy-distribution based local search to improve them. We evaluate the role of each operator of the proposed approach DBGSOM-NDE and compare its performance with a number of state-of-the-art niching techniques demonstrating its competitiveness and superiority, especially on high-dimensional and nonlinear problems taken from the existing literature. Finally, a hyper-parametric study is provided demonstrating weak dependence of them to the algorithm’s performance.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112862"},"PeriodicalIF":7.2,"publicationDate":"2025-02-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143437273","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Space-depth mutual compensation for fine-grained fabric defect detection model

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2025-02-17 DOI: 10.1016/j.asoc.2025.112869

Kailong Zhou, Jianhui Jia, Weitao Wu, Miao Qian, Zhong Xiang

In recent years, using the deep learning approach in the textile industry for defect detection has emerged as a prominent research. However, detecting fabric defects remains challenging due to the small size and small number of fabric defect features. Traditional down-sampling operations that result in loss of feature information, interpolation up-sampling operations that add a lot of background redundant information, and interference with fabric images from external sources such as lighting or electromagnetic devices are significant barriers to achieving accurate defect detection using existing methods. In this work, we introduced a lightweight fabric defect detection method with enhanced resistance to interference. Firstly, we use YOLOv7-tiny as the basic model and integrate the Spatial Pyramid Dilated Convolution (SPD) and Efficient Channel Attention (ECA) modules to enhance the original MP-1 and Effective Long-Range Aggregation Network (ELAN) modules to retain fine-grained information, solve the problem of down-sampled feature loss and improve feature importance allocation. Secondly, a distinctive up-sampling Module (DTS) was proposed to replace the traditional interpolation up-sampling. The module expands the feature map size without adding extraneous information, thus ensuring more efficient integration of features of different sizes. Finally, a novel noise filtering technique called the Color Space Iterative (CSI) method was proposed to filter noise interference quickly and conveniently. Experiments on the open-source DAGM and TILDA defect datasets, as well as supplementary tests on CIFAR10 datasets for the CSI method, have yielded promising results. With a mere 3.4M parameters, the proposed lightweight model underscores the method’s superiority over the baseline in balancing model parameters, detection speed, and accuracy.

{"title":"Space-depth mutual compensation for fine-grained fabric defect detection model","authors":"Kailong Zhou, Jianhui Jia, Weitao Wu, Miao Qian, Zhong Xiang","doi":"10.1016/j.asoc.2025.112869","DOIUrl":"10.1016/j.asoc.2025.112869","url":null,"abstract":"<div><div>In recent years, using the deep learning approach in the textile industry for defect detection has emerged as a prominent research. However, detecting fabric defects remains challenging due to the small size and small number of fabric defect features. Traditional down-sampling operations that result in loss of feature information, interpolation up-sampling operations that add a lot of background redundant information, and interference with fabric images from external sources such as lighting or electromagnetic devices are significant barriers to achieving accurate defect detection using existing methods. In this work, we introduced a lightweight fabric defect detection method with enhanced resistance to interference. Firstly, we use YOLOv7-tiny as the basic model and integrate the Spatial Pyramid Dilated Convolution (SPD) and Efficient Channel Attention (ECA) modules to enhance the original MP-1 and Effective Long-Range Aggregation Network (ELAN) modules to retain fine-grained information, solve the problem of down-sampled feature loss and improve feature importance allocation. Secondly, a distinctive up-sampling Module (DTS) was proposed to replace the traditional interpolation up-sampling. The module expands the feature map size without adding extraneous information, thus ensuring more efficient integration of features of different sizes. Finally, a novel noise filtering technique called the Color Space Iterative (CSI) method was proposed to filter noise interference quickly and conveniently. Experiments on the open-source DAGM and TILDA defect datasets, as well as supplementary tests on CIFAR10 datasets for the CSI method, have yielded promising results. With a mere 3.4M parameters, the proposed lightweight model underscores the method’s superiority over the baseline in balancing model parameters, detection speed, and accuracy.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112869"},"PeriodicalIF":7.2,"publicationDate":"2025-02-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143437274","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A seasonal-series LSTM network for irregular urban function zone recognition using Sentinel-2 images

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2025-02-15 DOI: 10.1016/j.asoc.2025.112876

Ting Hu , Mengyu Han , Zixuan Guo

Urban Functional Zone (UFZ) serves as the fundamental unit for urban planning and management, exerting a significant influence on the enhancement of urban administrative efficacy and the optimization of urban spatial configurations. The delineation of UFZs has benefited from the rich information and fine features provided by high spatial resolution (HSR) remote sensing images. It is recognized that the temporal dynamics of ground objects exhibit seasonal disparities across various UFZs. However, HSR images typically lack seasonal information and come with high acquisition costs. Therefore, this study introduces a novel classification framework for UFZ, leveraging the all-seasonal availability of Sentinel-2 remote sensing images. This framework is designed to capture the spectral, spatial, and temporal features intrinsic to UFZs, thereby enabling a detailed mapping of these zones. The proposed method is articulated in three sequential stages: Initially, to balance the significant scale difference of block (i.e., the fundamental mapping unit) size in 10-meter resolution remote sensing images, an adaptive gradient perception (AGP) mechanism is used to guide the feature extraction of different-scale blocks. Subsequently, the bag of visual words (BOVW) model is deployed to distill block-level spectral-spatial features. This is complemented by the introduction of a seasonal series LSTM network, engineered to apprehend block-level temporal dynamic, particularly focusing on the spectral-temporal signatures that distinguish different UFZs. The proposed framework is applied to UFZ classification in five typical cities in China. The resultant overall accuracy (OA) for all cases reaches around 93 %, marking a noteworthy improvement of approximately 7 % over existing methods. Our results demonstrate the superiority and portability of this framework, as well as the significant potential of open-source remote sensing images in large-scale UFZ mapping.

{"title":"A seasonal-series LSTM network for irregular urban function zone recognition using Sentinel-2 images","authors":"Ting Hu , Mengyu Han , Zixuan Guo","doi":"10.1016/j.asoc.2025.112876","DOIUrl":"10.1016/j.asoc.2025.112876","url":null,"abstract":"<div><div>Urban Functional Zone (UFZ) serves as the fundamental unit for urban planning and management, exerting a significant influence on the enhancement of urban administrative efficacy and the optimization of urban spatial configurations. The delineation of UFZs has benefited from the rich information and fine features provided by high spatial resolution (HSR) remote sensing images. It is recognized that the temporal dynamics of ground objects exhibit seasonal disparities across various UFZs. However, HSR images typically lack seasonal information and come with high acquisition costs. Therefore, this study introduces a novel classification framework for UFZ, leveraging the all-seasonal availability of Sentinel-2 remote sensing images. This framework is designed to capture the spectral, spatial, and temporal features intrinsic to UFZs, thereby enabling a detailed mapping of these zones. The proposed method is articulated in three sequential stages: Initially, to balance the significant scale difference of block (i.e., the fundamental mapping unit) size in 10-meter resolution remote sensing images, an adaptive gradient perception (AGP) mechanism is used to guide the feature extraction of different-scale blocks. Subsequently, the bag of visual words (BOVW) model is deployed to distill block-level spectral-spatial features. This is complemented by the introduction of a seasonal series LSTM network, engineered to apprehend block-level temporal dynamic, particularly focusing on the spectral-temporal signatures that distinguish different UFZs. The proposed framework is applied to UFZ classification in five typical cities in China. The resultant overall accuracy (OA) for all cases reaches around 93 %, marking a noteworthy improvement of approximately 7 % over existing methods. Our results demonstrate the superiority and portability of this framework, as well as the significant potential of open-source remote sensing images in large-scale UFZ mapping.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112876"},"PeriodicalIF":7.2,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143445321","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

GRU-ARX model-based adaptive error compensation predictive control strategy with application to quadrotor

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2025-02-15 DOI: 10.1016/j.asoc.2025.112829

Binbin Tian , Hui Peng , Zaihua Zhou

For a class of nonlinear dynamic systems, accurately characterizing the dynamic characteristics by building their physical models is still challenging. To deal with this issue, a novel deep learning network architecture, gated recurrent unit (GRU) neural network-based ARX model (GRU-ARX model), is developed in this study. In this model, the GRU network is executed to capture potential nonlinear mapping features of the system. And the pseudo linear ARX structure is adopted for making controller design easier, with the state-dependent parameters updated at each execution point. In view of this model, the model predictive control (MPC) algorithms for controlling the real nonlinear plant can be availably designed. However, faced with the appearance of sensibility with respect to internal or/and external factors in practical applications, the time-varying model may not perform well in control accuracy and robustness specification. Consequently, the operation of selecting the correction coefficients adaptively is combined with the MPC strategy to establish the adaptive MPC protocol focused on the error compensation, allowing for achieving the improved control accuracy and performance. Especially, the designed GRU-ARX model-based control algorithms, without and with the adaptive error compensation law are successfully applied to a practical quadrotor system, and the effectiveness of the accessed algorithms can be demonstrated by comparative results of real-time control experiments. These outcomes showcase that the proposed adaptive error compensation MPC algorithm exhibits superior control performance compared to other model-based controllers in trajectory tracking and anti-interference experiments, revealing its advantages over the traditional MPC algorithm.

{"title":"GRU-ARX model-based adaptive error compensation predictive control strategy with application to quadrotor","authors":"Binbin Tian , Hui Peng , Zaihua Zhou","doi":"10.1016/j.asoc.2025.112829","DOIUrl":"10.1016/j.asoc.2025.112829","url":null,"abstract":"<div><div>For a class of nonlinear dynamic systems, accurately characterizing the dynamic characteristics by building their physical models is still challenging. To deal with this issue, a novel deep learning network architecture, gated recurrent unit (GRU) neural network-based ARX model (GRU-ARX model), is developed in this study. In this model, the GRU network is executed to capture potential nonlinear mapping features of the system. And the pseudo linear ARX structure is adopted for making controller design easier, with the state-dependent parameters updated at each execution point. In view of this model, the model predictive control (MPC) algorithms for controlling the real nonlinear plant can be availably designed. However, faced with the appearance of sensibility with respect to internal or/and external factors in practical applications, the time-varying model may not perform well in control accuracy and robustness specification. Consequently, the operation of selecting the correction coefficients adaptively is combined with the MPC strategy to establish the adaptive MPC protocol focused on the error compensation, allowing for achieving the improved control accuracy and performance. Especially, the designed GRU-ARX model-based control algorithms, without and with the adaptive error compensation law are successfully applied to a practical quadrotor system, and the effectiveness of the accessed algorithms can be demonstrated by comparative results of real-time control experiments. These outcomes showcase that the proposed adaptive error compensation MPC algorithm exhibits superior control performance compared to other model-based controllers in trajectory tracking and anti-interference experiments, revealing its advantages over the traditional MPC algorithm.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112829"},"PeriodicalIF":7.2,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143455040","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

JaunENet: An effective non-invasive detection of multi-class jaundice deep learning method with limited labeled data

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2025-02-14 DOI: 10.1016/j.asoc.2025.112878

Yuanting Ma , Yu Meng , Xiaojun Li , Yutong Fu , Yan Xu , Yanfei Lu , Futian Weng

Jaundice, caused by elevated bilirubin levels, manifests as yellow discoloration of the eyes, mucous membranes, and skin, often serving as a clinical indicator of conditions such as hepatitis or liver cancer. This study introduces a non-invasive, multi-class jaundice detection framework that utilizes weakly supervised pre-training on large-scale medical images, followed by transfer learning and fine-tuning on 450 collected jaundice cases. Compared to existing studies, our classification approach is more detailed, encompassing a wider range of jaundice samples, including cases of occult jaundice, thereby enabling the accurate detection of more complex and subtle forms of the condition. Our model demonstrates exceptional performance on an independent test set, achieving an accuracy of 98.9 %, sensitivity of 0.991, specificity of 0.999, AUC of 0.999, and an F1-score of 0.990. Notably, the model’s computational efficiency is optimized for mobile deployment, requiring only 0.128 GFLOPs per image. Furthermore, the reliability of the model in identifying nuanced pathological features is validated through SHAP-based interpretability analyses. These findings highlight that weakly supervised pre-training outperforms methods reliant on detailed annotations, providing profound insights into small-sample deep learning applications in medical imaging and paving the way for more precise and scalable diagnostic tools.

{"title":"JaunENet: An effective non-invasive detection of multi-class jaundice deep learning method with limited labeled data","authors":"Yuanting Ma , Yu Meng , Xiaojun Li , Yutong Fu , Yan Xu , Yanfei Lu , Futian Weng","doi":"10.1016/j.asoc.2025.112878","DOIUrl":"10.1016/j.asoc.2025.112878","url":null,"abstract":"<div><div>Jaundice, caused by elevated bilirubin levels, manifests as yellow discoloration of the eyes, mucous membranes, and skin, often serving as a clinical indicator of conditions such as hepatitis or liver cancer. This study introduces a non-invasive, multi-class jaundice detection framework that utilizes weakly supervised pre-training on large-scale medical images, followed by transfer learning and fine-tuning on 450 collected jaundice cases. Compared to existing studies, our classification approach is more detailed, encompassing a wider range of jaundice samples, including cases of occult jaundice, thereby enabling the accurate detection of more complex and subtle forms of the condition. Our model demonstrates exceptional performance on an independent test set, achieving an accuracy of 98.9 %, sensitivity of 0.991, specificity of 0.999, AUC of 0.999, and an F1-score of 0.990. Notably, the model’s computational efficiency is optimized for mobile deployment, requiring only 0.128 GFLOPs per image. Furthermore, the reliability of the model in identifying nuanced pathological features is validated through SHAP-based interpretability analyses. These findings highlight that weakly supervised pre-training outperforms methods reliant on detailed annotations, providing profound insights into small-sample deep learning applications in medical imaging and paving the way for more precise and scalable diagnostic tools.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112878"},"PeriodicalIF":7.2,"publicationDate":"2025-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143437272","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Basic emotion detection accuracy using artificial intelligence approaches in facial emotions recognition system: A systematic review

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2025-02-11 DOI: 10.1016/j.asoc.2025.112867

Chia-Feng Hsu , Sriyani Padmalatha Konara Mudiyanselage , Rismia Agustina , Mei-Feng Lin

Facial emotion recognition (FER) systems are pivotal in advancing human communication by interpreting emotions such as happiness, sadness, anger, fear, surprise, and disgust through artificial intelligence (AI). This systematic review examines the accuracy of detecting basic emotions, evaluates the features, algorithms, and datasets used in FER systems, and proposes a taxonomy for their integration into healthcare. A comprehensive search of six databases, covering publications from January 1990 to March 2023, identified 4073 articles, with 35 studies meeting inclusion criteria.

The review revealed that happiness and surprise achieved the highest mean detection accuracies (96.42 % and 96.32 %, respectively), whereas anger and disgust exhibited lower accuracies (91.68 % and 93.71 %, respectively). Fear and sadness had a mean accuracy of 93.87 %. Among AI algorithms, GFFNN demonstrated the highest accuracy (100 %), followed by KNN (97.99 %) and DDBNN (97.77 %). CNN and SVM were the most commonly used algorithms, showing competitive accuracies. The CK+ dataset, while extensively employed, demonstrated a mean accuracy of 96.08 %, lower than RAVDESS, Oulu-CASIA, and other databases.

This taxonomy provides insights into FER systems' capabilities to enhance patient care by identifying emotional states, pain levels, and overall well-being. Future research should adopt diverse datasets and advanced algorithms to improve FER accuracy, enabling robust integration of these systems into healthcare practices.

{"title":"Basic emotion detection accuracy using artificial intelligence approaches in facial emotions recognition system: A systematic review","authors":"Chia-Feng Hsu , Sriyani Padmalatha Konara Mudiyanselage , Rismia Agustina , Mei-Feng Lin","doi":"10.1016/j.asoc.2025.112867","DOIUrl":"10.1016/j.asoc.2025.112867","url":null,"abstract":"<div><div>Facial emotion recognition (FER) systems are pivotal in advancing human communication by interpreting emotions such as happiness, sadness, anger, fear, surprise, and disgust through artificial intelligence (AI). This systematic review examines the accuracy of detecting basic emotions, evaluates the features, algorithms, and datasets used in FER systems, and proposes a taxonomy for their integration into healthcare. A comprehensive search of six databases, covering publications from January 1990 to March 2023, identified 4073 articles, with 35 studies meeting inclusion criteria.</div><div>The review revealed that happiness and surprise achieved the highest mean detection accuracies (96.42 % and 96.32 %, respectively), whereas anger and disgust exhibited lower accuracies (91.68 % and 93.71 %, respectively). Fear and sadness had a mean accuracy of 93.87 %. Among AI algorithms, GFFNN demonstrated the highest accuracy (100 %), followed by KNN (97.99 %) and DDBNN (97.77 %). CNN and SVM were the most commonly used algorithms, showing competitive accuracies. The CK+ dataset, while extensively employed, demonstrated a mean accuracy of 96.08 %, lower than RAVDESS, Oulu-CASIA, and other databases.</div><div>This taxonomy provides insights into FER systems' capabilities to enhance patient care by identifying emotional states, pain levels, and overall well-being. Future research should adopt diverse datasets and advanced algorithms to improve FER accuracy, enabling robust integration of these systems into healthcare practices.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112867"},"PeriodicalIF":7.2,"publicationDate":"2025-02-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143403525","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0