Next Article in Journal
Ultralong Carbon Nanotube Yarns Integrated as Electronic Functional Elements in Smart Textiles
Previous Article in Journal
Dyeing Performance of a Synthesized and Ultrafiltrated Bifunctional Reactive Dye with Two Vinylsulfone Groups on Cotton Fabrics
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Textiles
Volume 5
Issue 2
10.3390/textiles5020012
textiles-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

Artificial Intelligence Driving Innovation in Textile Defect Detection

by
Ahmet Ozek
1,
Mine Seckin
2,*,
Pinar Demircioglu
3 and
Ismail Bogrekci
3
1
Electrical-Electronics Engineering Department, Pamukkale University, Denizli 20100, Türkiye
2
Graduate School of Natural and Applied Sciences, Aydın Adnan Menderes University, Aydın 09100, Türkiye
3
Mechanical Engineering Department, Aydın Adnan Menderes University, Aydın 09100, Türkiye
*
Author to whom correspondence should be addressed.
Textiles 2025, 5(2), 12; https://doi.org/10.3390/textiles5020012
Submission received: 9 December 2024 / Revised: 13 March 2025 / Accepted: 17 March 2025 / Published: 4 April 2025
Browse Figures
Versions Notes

Abstract

:
The cornerstone of textile manufacturing lies in quality control, with the early detection of defects being crucial to ensuring product quality and sustaining a competitive edge. Traditional inspection methods, which predominantly depend on manual processes, are limited by human error and scalability challenges. Recent advancements in artificial intelligence (AI)—encompassing computer vision, image processing, and machine learning—have transformed defect detection, delivering improved accuracy, speed, and reliability. This article critically examines the evolution of defect detection methods in the textile industry, transitioning from traditional manual inspections to AI-driven automated systems. It delves into the types of defects occurring at various production stages, assesses the strengths and weaknesses of conventional and automated approaches, and underscores the pivotal role of deep learning models, especially Convolutional Neural Networks (CNNs), in achieving high precision in defect identification. Additionally, the integration of cutting-edge technologies, such as high-resolution cameras and real-time monitoring systems, into quality control processes is explored, highlighting their contributions to sustainability and cost-effectiveness. By addressing the challenges and opportunities these advancements present, this study serves as a comprehensive resource for researchers and industry professionals seeking to harness AI in optimizing textile production and quality assurance amidst the ongoing digital transformation.

Graphical Abstract

1. Introduction

The detection of faults in the textile industry is fundamental to quality control processes, with the early identification of defects during production playing a pivotal role in ensuring the final product’s quality [1]. Production defects can negatively affect the fabric’s aesthetic appearance and structural integrity, reducing the product’s value. Traditional manual inspection methods often lead to operator errors and low efficiency, while AI-powered automated approaches are transforming quality control processes by increasing accuracy and reducing costs [2]. Advanced technologies such as computer vision, image processing, and machine learning enable the rapid and reliable detection of fabric surface defects, production parameters, and pattern characteristics, thus raising quality standards in the textile sector [3,4]. These innovations contribute significantly to sustainability and competitiveness, allowing the industry to meet modern production expectations.
Defect detection processes have undergone a significant evolution historically, transitioning from traditional manual methods to modern technology [5]. In the early stages, these processes largely relied on visual inspection by operators, which was limited due to human errors and low efficiency. Towards the end of the 20th century, With the advancement of computing technology vision and image processing techniques, automated defect detection systems began to be used; however, these systems faced limitations such as low accuracy and limited computational capacity due to the technological constraints of that era [6]. In the 2000s, the increased availability of digital cameras, sensors, and computational power laid the foundation for the development of more precise and complex detection systems. Today, advanced technologies like artificial intelligence, machine learning, and deep learning have radically transformed defect detection processes, enabling rapid, accurate, and reliable identification of defects [3]. This technological transformation has not only raised quality control standards in the textile industry but also facilitated cost-effective and sustainable production goals [3,7,8,9].
However, machine learning-based defect detection methods are not solely limited to deep learning approaches such as Convolutional Neural Networks (CNNs). In the literature, traditional machine learning methods like Random Forests (RFs) and Markov Random Fields (MRFs) have also been widely used in textile defect detection, providing significant contributions to pattern recognition and defect localization [2,3]. These methods have shown effective results in identifying texture variations and classifying defect patterns in textile surfaces. This study aims to provide a broader perspective by examining both traditional and modern machine learning approaches, thereby highlighting the diversity of defect detection strategies in the textile industry.
This review aims to provide a comprehensive overview of research in areas such as defect detection, production optimization, and pattern segmentation within the textile industry. Unlike the existing literature, this review thoroughly examines current technological approaches and assesses their impact on textile processes. In addition to modern techniques such as artificial intelligence, machine learning, and deep learning, this research offers new perspectives on the limitations of traditional methods and their integration with contemporary technologies. The review summarizes the prominent methods in current research and elaborates on the challenges faced in defect detection processes, as well as the opportunities emerging to address these challenges. Additionally, the study discusses next-generation defect detection methodologies, including hybrid AI systems, real-time adaptive learning models, and predictive maintenance techniques, which are expected to shape the future of textile quality control. This study seeks to make significant scientific contributions to quality control and production efficiency in the textile sector and provides a comprehensive resource to guide the digital transformation of the textile industry.

2. Fault Types in Textiles

The textile industry involves a complex production process that spans several stages, from raw fiber to final yarn. Each stage contains various types of defects that directly affect product quality [10]. These defects can reduce production efficiency, increase product costs, and lead to customer dissatisfaction [11]. Errors occurring at every production stage such as fiber preparation, carding, spinning, weaving, and finishing, can not only affect the physical appearance of the fabric but also negatively impact its performance characteristics [12,13]. This article examines the most common types of defects encountered at different stages of the textile production process, addressing their causes, effects, and control methods. Proper identification and resolution of these defects are crucial for producing higher-quality textile products and play a significant role in enhancing competitiveness within the industry [3,14]. Table 1 provides a detailed summary of the types of defects encountered at each stage of textile production, along with explanations for each defect. These issues are factors that negatively impact the quality of the production process and must be identified and prevented through effective quality control methods. Accurate and timely detection of these defects is crucial for improving production efficiency and enhancing the quality of the final product (Figure 1).

3. Defect Detection Methods in Textiles

Quality control is considered one of the most critical components of production processes in the textile industry. Traditional methods focus on identifying production defects using manual and mechanical techniques. Among these, human inspection, mechanical systems, and optical sensors are the most commonly employed approaches. These methods have been utilized for decades, each with their own unique advantages and limitations. Below is a detailed analysis of these methods.

3.1. Traditional Human-Based Quality Control Inspection

Human inspection has been one of the most fundamental defect detection methods in textile production processes for many years [16]. In this approach, experienced operators and inspectors visually examine textile products to identify defects. Due to its flexibility and the operators’ familiarity with the production process, human inspection remains a preferred method [17]. However, as it relies heavily on human factors, this method has certain constraints. This section provides a detailed examination of the advantages and limitations of human inspection and the impact of operator dependence on quality control.
One of the primary advantages of human inspection is the ability of trained operators to quickly and effectively recognize various types of defects based on their experience. These operators possess extensive knowledge and intuitive understanding, enabling them to detect abnormalities in textile products. Human visual analysis skills and attention to detail are particularly crucial for identifying complex patterns and subtle defects. Experienced operators excel at detecting textural irregularities, color deviations, and pattern inconsistencies in fabrics [18,19].
However, there are significant limitations associated with human inspection. Operator fatigue, distractions, and individual perceptual differences can severely affect the quality of inspections. Prolonged work hours reduce operators’ focus, increasing the likelihood of errors. Moreover, since each individual’s perception ability varies, inconsistencies may arise during the inspection process. This inconsistency can create challenges in maintaining continuity and accuracy in quality control processes [20,21].
Continuous inspection by the human eye in large-scale production lines also presents challenges (Figure 2). Examining every product in detail becomes practically difficult in high-volume production, leading to a risk of increased defect rates [22]. Inconsistent operator performance may result in defective products entering the production stream. Consequently, while human inspection plays a significant role in quality control, it also carries various constraints and risks. Thus, integrating more automated and objective methods has become essential to overcome these challenges [23].

3.2. Mechanical and Optical Methods

Mechanical and optical methods are automated systems developed to reduce human errors in the textile industry. Mechanical methods monitor yarn breakages and physical defects in fabrics, enhancing the traceability of the production process. These systems continuously control the physical properties of the fabric, allowing for quick identification of abnormalities. In contrast, optical methods excel in detecting visual defects, such as color inconsistencies and pattern irregularities. Automated sensors and detection systems deliver faster and more reliable results, unaffected by human operators’ distractions. Their integration significantly improves quality control processes and elevates production standards [24,25].
To provide a clearer perspective, specific examples of existing commercial systems include the Uster Quantum 3 yarn clearer, which detects thin and thick faults in yarn, and the BarcoVision Cyclops system, which utilizes optical sensors to analyze fabric surface defects. These systems have been reported to improve defect detection rates by over 85% in industrial settings.
The combination of both systems provides a more comprehensive approach to quality control in textile production. While mechanical detection systems are highly efficient in identifying physical defects, optical systems quickly pinpoint aesthetic issues. However, each method has limitations: mechanical systems cannot detect visual flaws, and optical methods may be influenced by environmental factors. Consequently, the joint application of mechanical and optical methods allows for early detection of both physical and visual defects, preventing defective products from advancing through the production line and optimizing efficiency [26,27].
Furthermore, in practical applications, combined mechanical–optical setups are calibrated using benchmark defect samples under controlled lighting and tension conditions. These calibrations are critical for ensuring consistent performance in automated inspection systems.

3.3. Image Processing and Computer Vision

The textile industry demands fast, accurate, and automated defect detection methods to ensure the sustainability of high-quality production processes [28]. Traditional approaches, which often rely on manual inspections, are prone to human errors. In contrast, modern production processes aim to achieve higher efficiency and accuracy through computer-aided solutions (Figure 3). In this context, image processing and computer vision techniques have emerged as key technologies for the automatic detection of defects on textile surfaces.
The effectiveness of these methods depends on acquiring suitable data and accurately identifying defects. Critical to the data acquisition process are various camera systems that capture high-resolution images, enabling precise defect detection and classification. These systems, integrated with advanced algorithms, significantly enhance the reliability and consistency of quality control processes [5,29,30,31].
The detection of textile defects relies heavily on advanced camera systems that enable the visual inspection of fabrics during production. These systems offer tailored techniques to identify various defects based on their resolution, speed, and spectral capabilities. Standard cameras are widely used for basic quality control tasks due to their cost-effectiveness and sufficient performance in identifying large-scale defects like seam errors or significant yarn irregularities [32,33]. In contrast, high-resolution cameras provide detailed data, allowing the detection of minute flaws such as small holes, scratches, or broken threads, thus improving precision in quality control processes [34,35].
To address the challenges posed by high-speed production lines, high-speed cameras are employed. These systems are capable of monitoring rapidly moving fabric surfaces in real-time, effectively detecting transient defects and ensuring dynamic flaws are identified [36,37,38]. Additionally, hyperspectral cameras, with their ability to capture images across multiple spectral bands, analyze the chemical and structural properties of textiles in detail. They excel in detecting defects caused by material or color variations and can accurately examine fabric texture and dye quality [39,40,41].
Additionally, line-scan cameras are used for scanning continuously moving wide fabric surfaces and inspecting large areas at high resolution. These cameras scan the surface of moving fabric line by line, enabling precise monitoring of extensive surface areas and accurate localization of defects. Line-scan technology is particularly preferred for the quality control of large fabric rolls and wide textile products [42,43,44]. These camera systems enhance quality control performance in production processes, allowing for earlier defect detection and significantly contributing to improved production efficiency and product quality. These diverse camera systems directly influence the richness and quality of datasets used for the automated detection of defects on textile surfaces. Consequently, they improve the accuracy and performance of machine learning and deep learning algorithms developed for this purpose [45]. These datasets are created to identify and classify fabric defects, with labels based on geometric properties such as texture and patterns. Therefore, these data provide a solid foundation for developing computer vision-based defect detection systems in the textile industry.
The integration of high-resolution cameras and real-time monitoring systems into quality control processes in the textile industry significantly enhances production efficiency and product quality. High-resolution cameras meticulously scan fabric surfaces, enabling the detection of even the smallest defects. With their capability to identify fine issues such as color variations, linear weaving faults, and pattern misalignments with high precision, these cameras play a critical role in maintaining quality standards [8]. Real-time monitoring systems allow continuous inspection of fabric surfaces during production and enable the prompt identification of defects, facilitating immediate intervention without interrupting the production process. Supported by advanced image processing algorithms, these systems not only detect defects but also classify their types and dimensions, providing detailed defect analysis. This integration minimizes production errors and ensures consistent quality standards by detecting both visual and structural defects instantly. Additionally, the detailed data provided by these systems hold significant potential for continuous improvement in production processes and the establishment of data-driven decision-making mechanisms. In conclusion, the integration of high-resolution cameras and real-time monitoring systems into production lines offers substantial contributions to the development of higher-quality and more cost-efficient production processes in the textile industry [46].
Below is a table listing various datasets used for detecting textile defects and providing access to these datasets. In the design and development of quality control systems based on new technologies, the information provided by these datasets tailored to the textile industry is utilized extensively. Optimizing algorithms and visual inspection methods for identifying defects on fabric surfaces becomes possible through a detailed analysis of the examples contained in these datasets.
Datasets that include diverse fabric types, pattern varieties, defect types, and color variations play a crucial role in comparing algorithm performances and predicting the success of proposed methods in practical applications (Table 2). Furthermore, these databases offer researchers comprehensive references for defect localization, defect type classification, and evaluating the effectiveness of different imaging techniques (e.g., hyperspectral, high-speed, or high-resolution cameras). Thus, utilizing these datasets during the development of next-generation quality control systems contributes to creating more effective and robust methods, ultimately enhancing quality control efficiency in production processes.

3.3.1. Image Processing Techniques

Image processing is a technique that applies various mathematical and algorithmic operations to extract information from an image and improve its quality. Image processing techniques play a crucial role in textile defect detection. However, while high-resolution cameras enhance defect identification, they also introduce cost and computational challenges. A comparative analysis of cost vs. performance shows that high-resolution systems require approximately 30% higher computational resources than standard-resolution systems. In a production setting, a hyperspectral camera system can cost upwards of $50,000, whereas traditional optical cameras cost around $5000–$10,000.
The image processing techniques used for textile defect detection are generally examined in two main stages: preprocessing and feature extraction [9,59].
In the preprocessing stage, fabric images undergo operations such as noise reduction, brightness adjustments, and contrast enhancement to make them more suitable for defect detection. The improvements made at this stage enhance the image quality, allowing for more accurate defect detection [60].
The second stage, feature extraction, involves techniques such as edge detection, texture analysis, and color feature extraction to identify irregularities on the surface. Edge-based methods aim to detect errors on textile surfaces by identifying sudden changes in the regular structure of the fabric. Edge detection methods such as Canny edge detection and the Sobel filter capture sudden transition points on the fabric surface, marking regions where defects are present. These methods are particularly effective on fabrics with linear patterns [9].
On the other hand, texture-based methods focus on specific textural properties of textile surfaces and aim to detect errors in case of disruptions in these patterns. Techniques like Gabor filters, the Gray Level Co-occurrence Matrix (GLCM), and fractal analysis analyze the fabric’s textural characteristics to detect smoothness or irregularities on the surface. These methods are especially effective at identifying surface defects in fabrics with complex patterns. Thus, the combined use of edge-based and texture-based techniques enables successful detection of defects in textiles with different pattern structures [9,61].
The integration of high-resolution cameras and real-time monitoring systems offers significant advantages in textile manufacturing processes but also presents some challenges. One of the main challenges is the data processing load, which requires high computational power and advanced algorithms to process large volumes of data generated by high-resolution images in real time [54]. In high-speed production lines, it is crucial to optimize processing power and data processing workflows for real-time analysis of these data. Another significant challenge is the cost factor [55]. The cost of high-resolution cameras and the integrated monitoring systems in which these cameras are incorporated represents a substantial financial investment, especially in large-scale production facilities. In the future, there will be a need for AI-based optimization studies and the development of faster, lighter data processing algorithms to make these systems more efficient and cost-effective. Additionally, the use of real-time analysis platforms integrated with cloud-based monitoring systems has the potential to improve process efficiency and minimize production errors.

3.3.2. Defect Detection with Machine Learning

Traditional quality control methods largely rely on the human eye, creating a time-consuming and error-prone process. To overcome the limitations of these methods and provide more efficient production processes, machine learning (ML)-based approaches have rapidly gained popularity in the textile industry in recent years [3,62]. When integrated with image processing techniques, machine learning can automatically detect defects on fabric surfaces and classify these defects with high accuracy. Computer vision technology, in addition to image processing methods, uses artificial intelligence algorithms to make the process of extracting and analyzing meaningful information from images more efficient.
Among traditional machine learning methods, Support Vector Machines (SVMs) and Artificial Neural Networks (ANNs) stand out. These methods create classification models using training data to accurately identify defects on fabric surfaces [63,64].
Machine learning-based methods are typically used in three main stages for detecting textile defects: data collection, feature extraction, and classification. In the data collection phase, images of textile products are gathered using various camera systems, and these images, consisting of both defective and non-defective examples, provide training data for classification algorithms [3,65]. In the feature extraction phase, the prominent features on the fabric surface, such as texture patterns, color distributions, and designs, are analyzed to extract meaningful features that distinguish defects. Finally, in the classification phase, the extracted features are used to classify defective and non-defective areas on the fabric through machine learning algorithms, allowing defects to be correctly categorized and enabling the rapid detection of production errors.

3.3.3. Defect Detection with Deep Learning

In recent years, traditional quality control methods have been replaced by more precise and automated systems. In this context, deep learning-based algorithms, especially with their image processing and pattern recognition capabilities, offer a significant solution for detecting textile defects. Deep learning uncovers hidden patterns and relationships within complex data structures, enabling the detection of defects on textile surfaces with high accuracy. As a result, quality control processes become faster, more reliable, and consistent, minimizing human errors. Deep learning models have shown remarkable improvements in defect recognition rates, achieving accuracy levels ranging from 80% to 95% [2,7,58].
Deep learning-based approaches for detecting defects on textile surfaces, particularly Convolutional Neural Networks (CNNs), are attracting increasing attention from researchers and practitioners [66,67]. CNN models offer advanced analytical capabilities for detecting color, pattern, and texture irregularities on fabric surfaces, allowing defects to be identified with high precision [68]. These models, which analyze multidimensional patterns in fabric images, have the capacity to distinguish even the fine details that the human eye cannot perceive. CNN models trained on large datasets can quickly identify irregularities or color deviations in the production process by comparing normal and abnormal conditions [69]. Despite their high accuracy, the effectiveness of CNNs heavily depends on the size of training datasets. The literature suggests that a minimum of 10,000 labeled images is required for reliable deep learning model training in industrial textile defect detection. Smaller datasets risk model overfitting and reduced generalization to real production conditions. The application of such technology in the textile industry significantly contributes to improving production quality, reducing costs through the elimination of defective products, and increasing production efficiency [70]. It is also important to clarify whether the reported accuracy figures of “80% to 95%” originate from controlled lab experiments or real industrial settings. Many reported results in the literature are based on benchmark datasets under controlled conditions, which may not fully represent challenges encountered in industrial environments.
The integration of deep learning-based quality control technologies into the textile industry not only simplifies defect detection but also supports sustainability in production processes [70,71,72] (s. 4). Advanced defect detection systems contribute to preventing waste, while optimizing the use of energy and raw materials, thus reducing environmental impact [73]. Early detection of defects and the possibility for quick intervention help prevent production delays, resulting in labor and cost savings [71,73]. In this direction, deep learning applications encourage the adoption of an eco-friendly and competitive approach in the textile industry, contributing to a more sustainable production cycle through the integration of innovative technologies [74,75].

3.3.4. Random Forests (RFs) for Textile Defect Detection

Random Forests (RFs) is a supervised ensemble learning method that operates by generating numerous decision trees while training and producing the mode of the classifications for classification problems or the mean prediction for regression problems as output [76]. The algorithm reduces variance and prevents overfitting by averaging predictions from various trees. For textile use, RF is employed for recognizing surface defects by classifying feature vectors depending on various image features. These kinds of features are typically texture descriptors (for instance, Local Binary Patterns, Gabor filters), spectral features (for example, RGB/HSV histograms), and spatial details [77].
For example, Bourdeau-Laferrière et al. (2021) proposed an RF-based system for the optimization of quality control in textile dyeing processes [3,74]. The system used feature extraction of high-resolution images to identify color variations and imperfections and RF classification to identify anomalies and defects.
Random Forests (RFs) offer some technical advantages in textile defect inspection, e.g., low overfitting bias achieved by bootstrapped training sets and averaging output over ensembles of several decision trees to guarantee good generalization even in high-dimensional feature spaces [78,79]. RF can handle high-dimensional feature vectors of different image descriptors such as texture (Local Binary Patterns and Gabor filters), color (RGB and HSV histograms), and structural edge features (Canny and Sobel operators) without loss of performance. RF can cope with class-imbalanced datasets with high sensitivity and specificity by employing techniques such as Synthetic Minority Over-sampling (SMOTE) and class weighting. Another key benefit is its capacity for interpretation, as Random Forest (RF) has the capacity to assess feature importance, thereby enabling root cause analysis through the identification of essential defect-related features like texture discontinuities. Nonetheless, RF’s generalization capability might not be robust in relation to deep learning models, especially when there are enormous labeled datasets. Proper tuning of hyperparameters like the number of trees (n_estimators), maximum depth (max_depth), and the minimum number of samples per leaf node (min_samples_leaf) is crucial [80,81,82].

3.3.5. Textile Image Analysis with Markov Random Fields (MRFs)

Markov Random Fields (MRFs) are an effective statistical method that has been extensively applied in textile defect detection due to their capability in modeling spatial interdependencies among pixels and thus facilitating efficient texture and pattern analysis. Through the analysis of dependencies among a target pixel and its neighbors, MRF-based approaches have the capacity to maintain the integrity of texture and thus minimize false positives as well as enhance fault localization through the precise identification of individual faulty areas on the fabric surface [83]. Cai and Liu (2002) [84] and Jing et al. (2016) point out its excellence in pattern recognition and image segmentation of printed fabric. MRF’s multi-scale analysis capability allows it to detect defects at varied resolutions, therefore being suitable for images of various qualities [85]. However, its application is beset with challenges, e.g., high computational cost due to iterative optimization methods (e.g., belief propagation, graph cuts) and the challenge of tuning crucial parameters like neighborhood size and potential functions.

3.4. Classification and Application of AI Technologies in Textile Defect Detection

The application of artificial intelligence (AI) technologies in textile defect detection encompasses a wide array of approaches, each with unique strengths and limitations. For a clearer understanding and more structured analysis, the following classification of AI-based methods is presented:

3.4.1. Traditional Machine Learning Approaches

Traditional machine learning methods such as Support Vector Machines (SVMs), Random Forests (RFs), and Markov Random Fields (MRFs) are widely utilized for textile defect detection. These methods primarily rely on feature extraction techniques to analyze fabric surface characteristics and classify defects.
Support Vector Machines (SVMs): utilized to separate defective and non-defective fabric samples based on features like texture, color, and structural properties. SVM is particularly effective in high-dimensional spaces.
Random Forests (RFs): an ensemble learning technique that generates multiple decision trees and aggregates their outputs to enhance defect classification accuracy. It is known for its robustness and capacity to handle noisy data.
Markov Random Fields (MRFs): a statistical modeling approach that captures spatial dependencies among pixels, making it effective for identifying texture inconsistencies and localizing defects in patterned textiles.

3.4.2. Deep Learning Approaches

Deep learning techniques, particularly Convolutional Neural Networks (CNNs), have revolutionized defect detection in textiles by enabling automated feature learning from large datasets.
Convolutional Neural Networks (CNNs): CNNs automatically learn hierarchical features from fabric images, enabling the detection of subtle defects that may escape traditional methods. Their high accuracy and robustness have made CNNs the dominant approach in recent years.
Hybrid models: combining CNNs with traditional machine learning algorithms like SVM or RF can further improve detection performance. Hybrid systems leverage the strength of CNNs in feature extraction and the precision of traditional classifiers.

3.4.3. Hybrid and Adaptive AI Systems

The future of textile defect detection lies in hybrid and adaptive systems that integrate multiple AI approaches and dynamically adjust to varying production conditions.
Hybrid AI systems: These systems combine machine learning and deep learning models to address the limitations of each approach. For example, CNNs can detect complex patterns, while RF can refine the classification based on extracted features.
Adaptive learning systems: These systems continuously learn from new data, enabling real-time adaptation to changes in fabric types, production speeds, and defect patterns.

3.5. Application of AI Technologies in Textile Defect Detection

The practical application of AI technologies in textile quality control involves several steps, from data acquisition to real-time defect detection and process optimization.
Data acquisition: high-resolution cameras and line-scan systems capture detailed images of fabric surfaces. Image preprocessing techniques such as noise reduction and contrast enhancement improve image quality.
Feature extraction: texture analysis, edge detection, and color analysis techniques identify distinctive features indicative of defects.
Model training and classification: machine learning and deep learning models are trained on labeled datasets to classify fabric defects. Transfer learning techniques can accelerate model development by leveraging pre-trained networks.
Real-time monitoring: AI-powered systems enable real-time defect detection during production, minimizing downtime and reducing defect rates.
Process optimization: insights from defect detection data facilitate process adjustments, improving overall production efficiency and quality.

3.6. Comparative Evaluation of Defect Detection Methods

To provide a comprehensive perspective on the effectiveness of different defect detection methods, Table 3 compares traditional, machine learning, and deep learning approaches based on key performance criteria:

3.7. Challenges and Future Directions

While AI technologies offer significant improvements in textile defect detection, certain challenges remain.
High initial costs: implementation of high-resolution cameras and deep learning systems involves significant investment.
Data requirements: deep learning models require large, high-quality datasets for effective training.
Real-time processing: high-speed production lines demand computationally efficient algorithms.
Model generalization: AI models must be adaptable to different fabric types and defect patterns.
Future research should focus on developing cost-effective, adaptive AI systems capable of real-time analysis. Collaborative efforts between academia and industry can drive innovation and standardization in textile quality control. These additions aim to enhance the paper by providing a clear classification of AI technologies, a detailed explanation of their applications, and a comparative evaluation to guide future research and industrial practices.

4. Conclusions

The textile industry has undergone a paradigm shift in quality control, evolving from manual inspections to leveraging artificial intelligence, machine learning, and deep learning methodologies. Traditional defect detection methods, while foundational, face limitations in scalability, accuracy, and efficiency. The adoption of AI-enhanced techniques, particularly deep learning models like Convolutional Neural Networks (CNNs), has addressed these challenges by enabling precise, real-time defect detection across diverse textile production environments.
High-resolution and hyperspectral camera systems, coupled with advanced image processing algorithms, now form the backbone of automated quality control. These technologies not only improve detection accuracy but also facilitate sustainable production practices by reducing waste, optimizing resource use, and minimizing defects at early stages. For instance, AI-driven defect detection systems have been shown to reduce fabric waste by approximately 20% in large-scale production facilities.
However, the integration of these systems presents challenges, such as high initial investment costs and the computational demand of real-time data processing. Furthermore, potential limitations of deep learning models include dataset bias and the difficulty of adapting models to varying textile textures and colors across different production environments. Addressing these challenges requires ongoing improvements in domain adaptation techniques and diverse training datasets.
Future advancements should focus on developing cost-effective, scalable AI solutions that can adapt to varying textile types and production conditions. Enhanced data collection frameworks, combined with Industry 4.0 integration, will play a critical role in fostering predictive maintenance and smart factory systems. Additionally, incorporating standardization efforts into AI-driven textile inspection, such as compliance with ISO 18436-8 for condition monitoring [86] and ISO 9001 [87] for quality management, will further enhance the robustness and adoption of these technologies.
By embracing these innovations, the textile industry can achieve greater efficiency, improved product quality, and a significant reduction in environmental impact, paving the way for a more sustainable and competitive future.
In summary, the transition from conventional inspection to AI-driven quality control represents a transformative milestone for the textile industry. This evolution underscores the necessity of continuous technological adaptation to address both current and emerging challenges, ensuring the industry’s growth and alignment with global sustainability goals.
The transition from conventional inspection to AI-driven quality control represents a transformative milestone for the textile industry. Traditional defect detection methods, while foundational, face limitations in scalability, accuracy, and efficiency. Although deep learning-based methods, particularly Convolutional Neural Networks (CNNs), have become prominent in recent research, traditional machine learning methods such as Random Forests (RFs) and Markov Random Fields (MRFs) also offer significant contributions. These methods provide robust modeling, particularly in texture analysis and defect localization.
A key consideration in textile defect detection is the structural composition of different fiber types. Natural and synthetic fibers exhibit distinct defect patterns, necessitating tailored detection approaches. For instance, cotton fibers are prone to defects such as neps and short fibers due to their organic nature, while synthetic fibers like polyester often suffer from extrusion inconsistencies. Addressing these differences requires specialized image processing and machine learning techniques to effectively capture and analyze these variations.
However, the integration of these systems presents challenges, such as high initial investment costs, computational demand of real-time data processing, dataset bias, and the difficulty of adapting models to varying textile textures and colors across different production environments. Addressing these challenges requires ongoing improvements in domain adaptation techniques, diverse training datasets, and standardization efforts, such as compliance with ISO 18436-8 for condition monitoring and ISO 9001 for quality management.
Future research should focus on integrating AI with Industry 4.0, developing self-learning defect detection models, and utilizing synthetic data for robust AI training. By leveraging hybrid systems that combine RF, MRF, and CNN approaches, along with predictive maintenance and adaptive learning systems, the textile industry can further enhance the accuracy and efficiency of defect detection while reducing costs and environmental impact.
Advancements should prioritize scalable, cost-effective AI solutions capable of adapting to diverse textile types and production conditions. Enhanced data collection frameworks, combined with Industry 4.0 integration, will play a critical role in fostering predictive maintenance and smart factory systems. Embracing these innovations will enable the textile industry to achieve greater efficiency, improved product quality, and a significant reduction in environmental impact, paving the way for a more sustainable and competitive future.
This study examines traditional and artificial intelligence-based methods used in textile defect detection. Although deep learning-based methods have become prominent in recent research, traditional machine learning methods such as Random Forests (RFs) and Markov Random Fields (MRFs) also offer significant contributions. These methods provide robust modeling, particularly in texture analysis and defect localization. In the future, hybrid systems combining different approaches like RF, MRF, and CNN could be employed to improve the accuracy of defect detection systems. This would enable more sustainable and efficient quality control processes in the textile industry.

5. Future Trends and Improvement Strategies in Textile Defect Detection

The continuous evolution of artificial intelligence and machine learning technologies is shaping the future of textile defect detection. Despite the significant advancements in automated defect detection systems, there remain various challenges and opportunities for improvement in this field. This section discusses the emerging trends, future research directions, and potential enhancement strategies that can further optimize defect detection methodologies.

5.1. Integration of AI with Industry 4.0 and Smart Manufacturing

With the rapid advancement of Industry 4.0 technologies, textile defect detection is expected to become more integrated with smart manufacturing systems. The use of IoT-based real-time monitoring, cloud computing, and predictive analytics will allow textile manufacturers to proactively detect and prevent defects before they affect the production process. AI-driven smart systems can continuously learn from real-time data, adapting their defect detection algorithms based on dynamic production conditions.

5.2. Hybrid AI Models for More Accurate Defect Detection

Existing deep learning approaches primarily rely on supervised learning models, which require extensive labeled datasets for training. However, hybrid AI models, which combine traditional machine learning techniques (e.g., Random Forests and Support Vector Machines) with deep learning architectures, are emerging as promising solutions for enhancing defect detection performance. These hybrid models leverage the strengths of both rule-based decision systems and deep feature extraction, enabling more precise and adaptive defect classification.

5.3. Self-Learning and Adaptive AI Models

Traditional defect detection models are often static and require periodic retraining to maintain their accuracy. However, the future of textile quality control will likely involve self-learning AI systems that continuously improve through unsupervised and reinforcement learning methods. These models can adapt to changes in production conditions, new defect types, and variations in textile materials without requiring extensive human intervention.

5.4. Enhanced Data Augmentation and Synthetic Data for Model Training

One of the key challenges in AI-driven textile defect detection is the limited availability of high-quality labeled datasets. Future research will focus on data augmentation techniques, generative adversarial networks (GANs), and synthetic image generation to create more diverse and realistic defect images. This will help improve model robustness and generalizability across different textile patterns and defect types.

5.5. Sustainability and Energy-Efficient AI Models

As sustainability becomes a global priority, future textile defect detection systems will also need to align with eco-friendly and energy-efficient AI frameworks. Optimized AI algorithms with low computational power requirements will enable real-time defect detection with minimal energy consumption, reducing the carbon footprint of textile manufacturing.

Author Contributions

Conceptualization, A.O., M.S., P.D. and I.B.; methodology, A.O., M.S., P.D. and I.B.; software, A.O., M.S., P.D. and I.B.; validation, A.O., M.S., P.D. and I.B.; formal analysis, A.O., M.S., P.D. and I.B.; investigation, A.O., M.S., P.D. and I.B.; resources, A.O., M.S., P.D. and I.B.; data curation, A.O., M.S., P.D. and I.B.; writing—original draft preparation, A.O., M.S., P.D. and I.B.; writing—review and editing, A.O., M.S., P.D. and I.B.; visualization, A.O., M.S., P.D. and I.B.; supervision, A.O., M.S., P.D. and I.B.; project administration, A.O., M.S., P.D. and I.B.; funding acquisition, A.O., M.S., P.D. and I.B. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Seçkin, A.Ç.; Seckin, M. Detection of fabric defects with intertwined frame vector feature extraction. Alex. Eng. J. 2022, 61, 2887–2898. [Google Scholar]
  2. Datta, A.K.; Chandra, J.K. Detection of defects in fabric by morphological image processing. Woven Fabr. Eng. 2010, 217–232. [Google Scholar]
  3. Hariharan, V.; Sivaraman, K. Machine Learning Techniques for Quality Control in Textile Fabric Manufacturing. In Proceedings of the 2024 2nd International Conference on Sustainable Computing and Smart Systems (ICSCSS), Coimbatore, India, 10–12 July 2024; pp. 902–907. [Google Scholar]
  4. Bansode, P.; Goyal, P. Artificial intelligence in textile and fashion world. Int. J. Res.-Granthaalayah 2024, 12, 32–42. [Google Scholar] [CrossRef]
  5. Bandara, P.; Bandara, T.; Ranatunga, T.; Vimarshana, V.; Sooriyaarachchi, S.; Silva, C.D. Automated Fabric Defect Detection. In Proceedings of the 2018 18th International Conference on Advances in ICT for Emerging Regions (ICTer), Colombo, Sri Lanka, 26–29 September 2018; pp. 119–125. [Google Scholar]
  6. Rekha, M.N.; Prasad, M.L.; Mukherjee, S.; Nikam, S.V.; Sharma, S.; Reddy, P.C.S. An Automatic Error Recognition approach for Machine Translation Results based on Deep Learning. In Proceedings of the 2024 2nd International Conference on Computer, Communication and Control (IC4), Indore, India, 8–10 February 2024; pp. 1–8. [Google Scholar]
  7. Petrillo, A.; Rehman, M.; Baffo, I. Digital and Sustainable Transition in Textile Industry through Internet of Things Technologies: A Pakistani Case Study. Appl. Sci. 2024, 14, 5380. [Google Scholar] [CrossRef]
  8. Islam, M.M.; Mintoo, A.A.; Saimon, A.S.M. Enhancing textile quality control with iot sensors: A case study of automated defect detection. Int. J. Manag. Inf. Syst. Data Sci. 2024, 1, 19–30. [Google Scholar]
  9. Dhamal, K.S.; Joshi, C.; Chouhan, D.S. Fabric Defect Analysis in Textile Manufacturing: Evaluating Methods for Generic and Jacquard Fabrics. J. Propuls.Technol. 2023. [Google Scholar] [CrossRef]
  10. Hussain, T. Textile Processing Textile Engineering: An Introduction; Nawab, Y., Shaker, K., Eds.; De Gruyter Oldenbourg: Berlin, Germany; Boston, MA, USA, 2023; pp. 105–136. [Google Scholar]
  11. Kaka, S.; Maryam, M.; Wassan, R.K.; Sarwar, U.; Akbar, A.; Ansari, J. Reducing Outgoing Quality Level (OQL) in Textile Manufacturing Through Six Sigma Methodology: A Case Study. Available online: https://www.researchgate.net/profile/Kaka-Shuaib/publication/383032000_Reducing_Outgoing_Quality_Level_OQL_in_Textile_Manufacturing_Through_Six_Sigma_Methodology_A_Case_Study/links/66b8b63d299c327096c14f97/Reducing-Outgoing-Quality-Level-OQL-in-Textile-Manufacturing-Through-Six-Sigma-Methodology-A-Case-Study.pdf (accessed on 27 September 2024).
  12. Al-Laithy, A.H.A. The effect of mercerization process on physical, chemical and mechanical properties of compact cotton gassed yarns. Int. Des. J. 2019, 9, 37–52. [Google Scholar]
  13. Al-Laithy, A.H.A.; Abdel Al, E.A.B. The effect of different methods of weft yarn insertion in modern weaving looms on both functional and aesthetic properties of ladies summer fabrics. Int. Des. J. 2019, 9, 133–144. [Google Scholar]
  14. Loganathan, V.; Suganya, K. A Fabric and Production Defect Detection in the Apparel Industry Using DLT. In Disruptive Technologies for Sustainable Development; CRC Press: Boca Raton, FL, USA, 2024; pp. 58–62. [Google Scholar]
  15. Fiber Defect Detection of Inhomogeneous Voluminous Textiles|SpringerLink. Available online: https://link.springer.com/chapter/10.1007/978-3-319-59226-8_27 (accessed on 16 September 2024).
  16. Yazdi, H.R.; King, T.G. Application of ‘vision in the loop’for inspection of lace fabric. Real-Time Imaging 1998, 4, 317–332. [Google Scholar]
  17. Shi, M.; Fu, R.; Guo, Y.; Bai, S.; Xu, B. Fabric defect detection using local contrast deviations. Multimed. Tools Appl. 2011, 52, 147–157. [Google Scholar] [CrossRef]
  18. Ramdass, K.R.; Mokgohloa, K. An exploration of effective supervisory engagement in the clothing industry through activity sampling technique. S. Afr. J. Ind. Eng. 2022, 33, 165–176. [Google Scholar] [CrossRef]
  19. Braun, M.; Greve, M.; Brendel, A.B.; Kolbe, L.M. Humans Supervising Artificial Intelligence–Investigation of Designs to Optimize Error Detection. J. Decis. Syst. 2023, 33, 674–699. [Google Scholar] [CrossRef]
  20. Jain, D. Human resource issues and challenges prevailing in Indian textile & garment industry. Int. J. Home Sci. 2022, 8, 227–232. [Google Scholar]
  21. Negara, N.L.G.A.M.; Paramurthi, I.P.; Dewi, N.K.P.P. Effect of Occupational Safety and Health Procedures on Fabric Dyeing Process in Reducing Worker Fatigue at the TBS Textile Factory. Occup. Saf. Health 2023, 12, 167–175. [Google Scholar]
  22. Mahajan, P.M.; Kolhe, S.R.; Patil, P.M. A review of automatic fabric defect detection techniques. Adv. Comput. Res. 2009, 1, 18–29. [Google Scholar]
  23. Rožanec, J.M.; Montini, E.; Cutrona, V.; Papamartzivanos, D.; Klemenčič, T.; Fortuna, B.; Mladenić, D.; Veliou, E.; Giannetsos, T.; Emmanouilidis, C. Human in the AI Loop via xAI and Active Learning for Visual Inspection. In Artificial Intelligence in Manufacturing; Soldatos, J., Ed.; Springer Nature Switzerland: Cham, Switzerland, 2024; pp. 381–406. ISBN 978-3-031-46451-5. [Google Scholar]
  24. Lane, J.S. Textile Fabric Inspection System. 1998. Available online: https://patents.google.com/patent/US5774177A/en (accessed on 5 October 2024).
  25. Rahimunnisa, K. Textile Fabric Defect Detection. J. Innov. Image Process. 2022, 4, 165–172. [Google Scholar]
  26. Mefford, D.K.; Smith, R.A.; Lowell, J.R. Optical Detection System and Method 2020. Available online: https://patents.google.com/patent/US10809186B2/en (accessed on 5 October 2024).
  27. Mak, K.L.; Peng, P.; Lau, H.Y.K. A real-time computer vision system for detecting defects in textile fabrics. In Proceedings of the 2005 IEEE International Conference on Industrial Technology, Hong Kong, China, 14–17 December 2005; pp. 469–474. [Google Scholar]
  28. Mei, S.; Wang, Y.; Wen, G. Automatic Fabric Defect Detection with a Multi-Scale Convolutional Denoising Autoencoder Network Model. Sensors 2018, 18, 1064. [Google Scholar] [CrossRef]
  29. Habib, M.T.; Faisal, R.H.; Rokonuzzaman, M.; Ahmed, F. Automated Fabric Defect Inspection: A Survey of Classifiers. IJFCST 2014, 4, 17–25. [Google Scholar] [CrossRef]
  30. Jiang, J.L.; Wong, W.K. 1—Fundamentals of common computer vision techniques for textile quality control. In Applications of Computer Vision in Fashion and Textiles; Wong, W.K., Ed.; The Textile Institute Book Series; Woodhead Publishing: Sawston, UK, 2018; pp. 3–15. ISBN 978-0-08-101217-8. [Google Scholar]
  31. Liu, Q.; Wang, C.; Li, Y.; Gao, M.; Li, J. A fabric defect detection method based on deep learning. IEEE Access 2022, 10, 4284–4296. [Google Scholar]
  32. Roberts, J.W.; Elias, J.G.; Jullien, G.A. High Speed Defect Detection Apparatus Having Defect Detection Circuits Mounted in the Camera Housing. U.S. Patent No. 5,440,648, 8 August 1995. [Google Scholar]
  33. Tao, Y. Defective Object Inspection and Removal Systems and Methods for Identifying and Removing Defective Objects. U.S. Patent 5,960,098, 28 September 1999. [Google Scholar]
  34. Gehrig, D.; Scaramuzza, D. Are high-resolution event cameras really needed? arXiv 2022, arXiv:2203.14672. [Google Scholar]
  35. Bakhtiarnia, A.; Zhang, Q.; Iosifidis, A. Efficient High-Resolution Deep Learning: A Survey. ACM Comput. Surv. 2024, 56, 1–35. [Google Scholar] [CrossRef]
  36. Wang, K.-J.; Lee, Y.-X. Measuring defects in high-speed production lines—A three-phase convolutional neural network model. Meas. Sci. Technol. 2023, 34, 105903. [Google Scholar]
  37. Wibowo, A.; Setiawan, J.D.; Afrisal, H.; Mertha, A.A.S.M.M.J.; Santosa, S.P.; Wisnu, K.B.; Mardiyoto, A.; Nurrakhman, H.; Kartiwa, B.; Caesarendra, W. Optimization of Computational Resources for Real-Time Product Quality Assessment Using Deep Learning and Multiple High Frame Rate Camera Sensors. Appl. Syst. Innov. 2023, 6, 25. [Google Scholar] [CrossRef]
  38. Kłosowski, P.; Michałowska, K.; Przyborski, M.; Jurczak, W. Testing of Technical Fabrics under Fast Camera Control. Adv. Sci. Technology. Res. J. 2023, 17, 332–334. [Google Scholar]
  39. Clark, J.; Johnson, G.; Duran, O.; Argyriou, V. Fabric composition classification using hyper-spectral imaging. In Proceedings of the 2023 19th International Conference on Distributed Computing in Smart Systems and the Internet of Things (DCOSS-IoT), Pafos, Cyprus, 19–21 June 2023; pp. 347–353. [Google Scholar]
  40. Gbelec, J.; Sitko, R.; Saloň, Š.; Forgáč, R.; Očkay, M. The Use Of Hyperspectral Sensors In Aerial Inspections Of Overhead Transmission Lines. In Proceedings of the 2023 23rd International Scientific Conference on Electric Power Engineering (EPE), Brno, Czech Republic, 24–26 May 2023; pp. 1–4. [Google Scholar]
  41. Vlachou-Mogire, C.; Danskin, J.; Gilchrist, J.R.; Hallett, K. Mapping materials and dyes on historic tapestries using hyperspectral imaging. Heritage 2023, 6, 3159–3182. [Google Scholar] [CrossRef]
  42. Usamentiaga, R. Static calibration for line-scan cameras based on a novel calibration target. IEEE Trans. Instrum. Meas. 2022, 71, 1–12. [Google Scholar]
  43. Huo, S.; Zhang, B.; Muddassir, M.; Chik, D.T.; Navarro-Alarcon, D. A sensor-based robotic line scan system with adaptive ROI for inspection of defects over convex free-form specular surfaces. IEEE Sens. J. 2021, 22, 2782–2792. [Google Scholar]
  44. Yuan, G.; Zheng, L.; Ding, Y.; Zhang, H.; Zhang, X.; Liu, X.; Sun, J. A precise calibration method for line scan cameras. IEEE Trans. Instrum. Meas. 2021, 70, 1–9. [Google Scholar]
  45. Silvestre-Blanes, J.; Albero-Albero, T.; Miralles, I.; Pérez-Llorens, R.; Moreno, J. A Public Fabric Database for Defect Detection Methods and Results. Autex Res. J. 2019, 19, 363–374. [Google Scholar]
  46. Toigo, S.; Kasi, B.; Fornasier, D.; Cenedese, A. Flexible machine/deep learning microservice architecture for industrial vision-based quality control on a low-cost device. J. Electron. Imaging 2024, 33, 031208. [Google Scholar]
  47. By Aitex Fabric Image Database—Aitex 2023. Available online: https://www.aitex.es/afid/ (accessed on 6 October 2024).
  48. Fabric Defect Dataset. Available online: https://www.kaggle.com/datasets/rmshashi/fabric-defect-dataset (accessed on 8 December 2024).
  49. Singh, M. msminhas93/FabricDefect. 2024. Available online: https://github.com/msminhas93/FabricDefect (accessed on 8 December 2024).
  50. Fabric Defect Detection Dataset > Overview. Available online: https://universe.roboflow.com/muhammad-saad/fabric-defect-detection-8ndhb (accessed on 8 December 2024).
  51. Carrilho, R.; Hambarde, K.A.; Proença, H. A Novel Dataset for Fabric Defect Detection: Bridging Gaps in Anomaly Detection. Appl. Sci. 2024, 14, 5298. [Google Scholar] [CrossRef]
  52. Hambarde, K.; Silahtaroğlu, G.; Khamitkar, S.; Bhalchandra, P.; Shaikh, H.; Kulkarni, G.; Tamsekar, P.; Samale, P. Data Analytics Implemented over E-commerce Data to Evaluate Performance of Supervised Learning Approaches in Relation to Customer Behavior. In Soft Computing for Problem Solving; Das, K.N., Bansal, J.C., Deep, K., Nagar, A.K., Pathipooranam, P., Naidu, R.C., Eds.; Advances in Intelligent Systems and Computing; Springer: Singapore, 2020; Volume 1048, pp. 285–293. ISBN 978-981-15-0034-3. [Google Scholar]
  53. Zhang, C.; Feng, S.; Wang, X.; Wang, Y. ZJU-Leaper: A Benchmark Dataset for Fabric Defect Detection and a Comparative Study. IEEE Trans. Artif. Intell. 2020, 1, 219–232. [Google Scholar] [CrossRef]
  54. Computer Vision Group, Freiburg. Available online: https://lmb.informatik.uni-freiburg.de/resources/datasets/tilda.en.html (accessed on 8 December 2024).
  55. Bhutra, H. The-h-b/TILDA_400; 2023. Available online: https://github.com/the-h-b/TILDA_400 (accessed on 8 December 2024).
  56. Papers with Code-DAGM2007 Dataset. Available online: https://paperswithcode.com/dataset/dagm2007 (accessed on 8 December 2024).
  57. Bergmann, P.; Batzner, K.; Fauser, M.; Sattlegger, D.; Steger, C. The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection. Int. J. Comput. Vis. 2021, 129, 1038–1059. [Google Scholar] [CrossRef]
  58. Dorafshan, S.; Thomas, R.J.; Maguire, M. SDNET2018: An annotated image dataset for non-contact concrete crack detection using deep convolutional neural networks. Data Brief. 2018, 21, 1664–1668. [Google Scholar] [CrossRef]
  59. Shang, H.; Li, P.; Zhang, X.; Peng, X. Overview of fabric defect detection techniques based on computer vision. In Proceedings of the Third International Conference on Computer Graphics, Image, and Virtualization (ICCGIV 2023), Xiamen, China, 9–11 June 2023; Volume 12934, pp. 108–115. [Google Scholar]
  60. Mahmood, T.; Ashraf, R.; Faisal, C.M.N. An efficient scheme for the detection of defective parts in fabric images using image processing. J. Text. Inst. 2023, 114, 1041–1049. [Google Scholar] [CrossRef]
  61. Sharma, R.S.; Kuppili, V. A Novel Approach for Texture Defect Detection using Statistical Features and a Bio-Inspired Algorithm. In Proceedings of the Companion Conference on Genetic and Evolutionary Computation; ACM: Lisbon, Portugal, 2023; pp. 103–106. [Google Scholar]
  62. Kins, R.; Möbitz, C.; Gries, T. Towards autonomous learning and optimisation in textile production: Data-driven simulation approach for optimiser validation. J. Intell. Manuf. 2024, 1–26. [Google Scholar] [CrossRef]
  63. Hemalatha, R.; Muthumeenakshi, K.; Radha, S. A neural-network-based machine-learning model for fabric defect detection and classification using fused global features. Aust. J. Electr. Electron. Eng. 2023, 20, 371–386. [Google Scholar] [CrossRef]
  64. Yaşar Çıklaçandır, F.G.; Utku, S.; Özdemir, H. Determination of various fabric defects using different machine learning techniques. J. Text. Inst. 2024, 115, 733–743. [Google Scholar] [CrossRef]
  65. Barzizza, E.; Biasetton, N.; Ceccato, R.; Molena, A. Machine learning-based decision-making approach for predicting defects detection: A case study. IAES Int. J. Artif. Intell. (IJ-AI) 2024, 13, 3052–3060. [Google Scholar] [CrossRef]
  66. Sun, G.; Zhou, Z.; Gao, Y.; Xu, Y.; Xu, L.; Lin, S. A Fast Fabric Defect Detection Framework for Multi-Layer Convolutional Neural Network Based on Histogram Back-Projection. IEICE Trans. Inf. Syst. 2019, 102, 2504–2514. [Google Scholar]
  67. Li, C.; Li, J.; Li, Y.; He, L.; Fu, X.; Chen, J. Fabric Defect Detection in Textile Manufacturing: A Survey of the State of the Art. Secur. Commun. Netw. 2021, 2021, e9948808. [Google Scholar] [CrossRef]
  68. Zhao, S.; Yin, L.; Zhang, J.; Wang, J.; Zhong, R. Real-time fabric defect detection based on multi-scale convolutional neural network. IET Collab. Intell. Manuf. 2020, 2, 189–196. [Google Scholar] [CrossRef]
  69. Li, X.; Zhu, Y. A real-time and accurate convolutional neural network for fabric defect detection. Complex. Intell. Syst. 2024, 10, 3371–3387. [Google Scholar] [CrossRef]
  70. Li, F.; Xiao, K.; Hu, Z.; Zhang, G. Fabric defect detection algorithm based on improved YOLOv5. Vis. Comput. 2024, 40, 2309–2324. [Google Scholar] [CrossRef]
  71. Hassan, S.A.; Beliatis, M.J.; Radziwon, A.; Menciassi, A.; Oddo, C.M. Textile Fabric Defect Detection Using Enhanced Deep Convolutional Neural Network with Safe Human–Robot Collaborative Interaction. Electronics 2024, 13, 4314. [Google Scholar] [CrossRef]
  72. Kailasam, K.; Singh, J.; Gopika, G.S.; Sreekrishna, S. Fabric Defect Detection using Deep Learning. In Proceedings of the 2024 3rd International Conference on Sentiment Analysis and Deep Learning (ICSADL), Online, 13–14 March 2024; pp. 399–405. [Google Scholar]
  73. Nasim, M.; Mumtaz, R.; Ahmad, M.; Ali, A. Fabric Defect Detection in Real World Manufacturing Using Deep Learning. Information 2024, 15, 476. [Google Scholar] [CrossRef]
  74. Ramos, L.; Rivas-Echeverría, F.; Pérez, A.G.; Casas, E. Artificial intelligence and sustainability in the fashion industry: A review from 2010 to 2022. SN Appl. Sci. 2023, 5, 387. [Google Scholar] [CrossRef]
  75. Yılmaz, K.; Aksu, İ.Ö.; Göçken, M.; Demirdelen, T. Sustainable Textile Manufacturing with Revolutionizing Textile Dyeing: Deep Learning-Based, for Energy Efficiency and Environmental-Impact Reduction, Pioneering Green Practices for a Sustainable Future. Sustainability 2024, 16, 8152. [Google Scholar] [CrossRef]
  76. Janapareddy, D.; Yenduri, N.C. Credit Card Approval Prediction: A Comparative Analysis Between Logistic Regressionclassifier, Random Forest Classifier, Support Vectorclassifier with Ensemble Bagging Classifier. 2023. Available online: https://www.diva-portal.org/smash/record.jsf?pid=diva2:1784289 (accessed on 8 December 2024).
  77. Deotale, N.T.; Sarode, T.K. Fabric Defect Detection Adopting Combined GLCM, Gabor Wavelet Features and Random Decision Forest. 3D Res. 2019, 10, 5. [Google Scholar] [CrossRef]
  78. Jamunadevi, C.; Sathishkumar, V.E.; Prasath, S.; Sadhanayaki, S.; Jaikishore, N. Optimized Fabric Imperfection Detection via Image Analysis and Isolation Forest Technique. In Proceedings of the 2024 International Conference on Signal Processing, Computation, Electronics, Power and Telecommunication (IConSCEPT), Karaikal, India, 4–5 July 2024; pp. 1–6. [Google Scholar]
  79. Miao, Y.; Xu, Y. Random Forest-Based Analysis of Variability in Feature Impacts. In Proceedings of the 2024 IEEE 2nd International Conference on Image Processing and Computer Applications (ICIPCA), Shenyang, China, 28–30 June 2024; pp. 1130–1135. [Google Scholar]
  80. Li, X.; Fan, X.; Li, Q.; Zhao, X. RS-Net: Hyperspectral Image Land Cover Classification Based on Spectral Imager Combined with Random Forest Algorithm. Electronics 2024, 13, 4046. [Google Scholar] [CrossRef]
  81. Kim, Y.; Kim, S.Y.; Kim, H. Heterogeneous Random Forest. arXiv 2024, arXiv:2410.19022. [Google Scholar] [CrossRef]
  82. Salah, A.; Yevick, D. A Random Forest Model for Predicting and Analyzing the Performance of CNT TFET with Highly Doped Pockets. Advcd Theory Sims 2025, 8, 2400607. [Google Scholar] [CrossRef]
  83. Freguglia, V.; Garcia, N.L. Inference tools for Markov Random Fields on lattices: The R package mrf2d. J. Stat. Softw. 2022, 101, 1–36. [Google Scholar] [CrossRef]
  84. Cai, J.; Liu, Z.-Q. Pattern recognition using Markov random field models. Pattern Recognit. 2002, 35, 725–733. [Google Scholar] [CrossRef]
  85. Jing, J.; Li, Q.; Li, P.; Zhang, H.; Zhang, L. Image Segmentation of Printed Fabrics with Hierarchical Improved Markov Random Field in the Wavelet Domain. J. Eng. Fibers Fabr. 2016, 11, 155892501601100320. [Google Scholar] [CrossRef]
  86. ISO 18436-8:2013; Condition Monitoring and Diagnostics of Machines—Requirements for Qualification and Assessment of Personnel. ISO: Geneva, Switzerland, 2013. Available online: https://www.iso.org/standard/50608.html (accessed on 16 March 2025).
  87. ISO 9001:2015; Quality Management Systems—Requirements. ISO: Geneva, Switzerland, 2015. Available online: https://www.iso.org/standard/62085.html (accessed on 16 March 2025).
Textiles 05 00012 g001
Figure 1. Fiber defect detection of inhomogeneous voluminous textiles [15].
Figure 1. Fiber defect detection of inhomogeneous voluminous textiles [15].
Textiles 05 00012 g001
Textiles 05 00012 g002
Figure 2. Traditional human-supervised fabric quality control process, where an operator manually inspects textile defects under controlled lighting conditions. The manual approach is prone to human error and inefficiencies, which AI-based solutions aim to overcome.
Figure 2. Traditional human-supervised fabric quality control process, where an operator manually inspects textile defects under controlled lighting conditions. The manual approach is prone to human error and inefficiencies, which AI-based solutions aim to overcome.
Textiles 05 00012 g002
Textiles 05 00012 g003
Figure 3. Fabric defect detection by (a) manual and (b) automated optical inspection methods [28].
Figure 3. Fabric defect detection by (a) manual and (b) automated optical inspection methods [28].
Textiles 05 00012 g003
Table 1. Types of defects encountered at each stage of textile production and their explanations.
Table 1. Types of defects encountered at each stage of textile production and their explanations.
Production StageDefect TypeDescription
Fiber DefectsShort Fiber RatioFibers are shorter than the standard length.
Fiber Color VariationsInconsistent color tones among fibers.
Foreign Fiber ContaminationMixing of different types or colors of fibers.
Dirty FiberPresence of foreign materials or dust in fibers.
Decayed Fiber Fiber is weak or deteriorated in structure.
Carding and Combing DefectsKnotty FiberFiber is not properly separated, causing knots.
Insufficient CardingFiber is not adequately opened.
Incorrect Blend RatiosIncorrect proportion of fiber blends.
Roving DefectsIrregular ThicknessInconsistent thickness of the roving.
Formation of Neps (Small Knots) Formation of small knots or lumps.
Blending Errors Incorrect ratio of blended fibers.
Yarn DefectsIrregularity (Uster Fault)Irregularities in yarn thickness.
NepsFormation of small lumps or knots in the yarn.
Foreign Fiber ContaminationPresence of noticeable foreign fibers in the yarn.
Thin and Thick SpotsInconsistent thickness and thin areas in the yarn.
Weak Yarn StrengthYarn is weaker than expected.
Yarn Breakage Breaks in the yarn at certain points.
Twist ErrorsIncorrect number of twists in the yarn.
Spiral Spiral formation in yarn due to uneven twist.
Bobbin and Spool DefectsKnot ConnectionsIncorrect yarn connections, resulting in visible knots.
Uneven Winding Uneven winding of bobbin or spool causing fluffing.
Foreign MaterialPresence of dust, fiber, or other foreign materials in bobbin or spool.
Weaving Preparation DefectsWarp and Weft MissesMissing or extra warp and weft threads in the weave pattern.
Warp Breakage Broken warp threads cause gaps in the fabric.
Incorrect Pattern WeavingIncorrect or incomplete weaving of the pattern.
Warp Lines Visible lines created by warp threads in the fabric.
Weft Lines Visible lines created by weft threads in the fabric.
Knitting (Sweater) DefectsMissed LoopImproper formation of loops in knitting.
Unraveling Loops unraveling, causing cascading errors.
Pattern ErrorsIncorrect or faulty repetition of patterns.
Dyeing and Finishing DefectsColor InconsistencyColor variations between fabric batches.
StainingPresence of dye stains or foreign material marks on fabric.
FadingFading of color due to washing, light exposure, or friction.
Excessive Shrinkage or WrinklingFabric shrinks or wrinkles more than expected.
Printing DefectsMisaligned Pattern Incorrect placement of printed design.
Color Mismatch Inconsistencies in the colors used in printing.
Incomplete PrintDesign is not fully printed or incorrect design is used.
Physical DefectsHoles and CutsPresence of holes or cuts in the fabric.
Surface DefectsFabric surface exhibiting roughness, protruding threads, or irregularities.
Table 2. Textile defect datasets.
Table 2. Textile defect datasets.
RefSample SizeDefect
Types		Dataset Type
AITEX AFID [47]2457 types4096 × 256 gray images and segmentation mask
Fabric Defect Dataset [48]Not specifiedHorizontal, vertical, and hole defects640 × 360
FabricDefect [49]24Not Specified512 × 512 mas
Fabric Defect Detection [50]108Hole, Knot, Stain
Lusitano Dataset [51]33,684Various textile defects (tears, stains, knots, pattern errors)The training set contains 32 k normal images, while the testing set contains 1646 defect images and 1038 normal images [52].
ZJU-Leaper [53]2000Tears, stains, knots,Various textile patterns and defect images
TILDA Dataset [54]32007 defect classes (e.g., tears, stains, knots)768 × 512 pixel, 8-bit grayscale TIF images; each class contains 50 images
ZJU-Leaper GitHub Various textile defects such as tears, stains, knots, and pattern errorsVarious textile patterns and defect images
TILDA_400 [55]400Defects in different weaving patterns and textures64 × 64 pixel patch images
DAGM2007 [56]Not specifiedProduction-related defectsNot specified
MVTec AD [57]Not specifiedProduction defects such as stains, tears, and texture abnormalitiesHigh-resolution industrial images; various object and surface defects
USU Dataset [58]Not specifiedCracks, holes, irregularitiesNot specified
Table 3. Comparison of textile defect detection methods.
Table 3. Comparison of textile defect detection methods.
MethodAccuracySpeedAdaptabilityCostComplexity
Human InspectionLowLowHighLowLow
Mechanical/Optical SystemsMediumHighLowHighMedium
SVMHighMediumMediumMediumMedium
Random Forests (RFs)HighMediumHighMediumMedium
Markov Random Fields (MRFs)HighLowLowMediumHigh
CNNVery HighHighHighHighHigh
Hybrid SystemsVery HighHighVery HighHighVery High
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Ozek, A.; Seckin, M.; Demircioglu, P.; Bogrekci, I. Artificial Intelligence Driving Innovation in Textile Defect Detection. Textiles 2025, 5, 12. https://doi.org/10.3390/textiles5020012

AMA Style

Ozek A, Seckin M, Demircioglu P, Bogrekci I. Artificial Intelligence Driving Innovation in Textile Defect Detection. Textiles. 2025; 5(2):12. https://doi.org/10.3390/textiles5020012

Chicago/Turabian Style

Ozek, Ahmet, Mine Seckin, Pinar Demircioglu, and Ismail Bogrekci. 2025. "Artificial Intelligence Driving Innovation in Textile Defect Detection" Textiles 5, no. 2: 12. https://doi.org/10.3390/textiles5020012

APA Style

Ozek, A., Seckin, M., Demircioglu, P., & Bogrekci, I. (2025). Artificial Intelligence Driving Innovation in Textile Defect Detection. Textiles, 5(2), 12. https://doi.org/10.3390/textiles5020012

Article Metrics

Back to TopTop