1. The importance of fabric quality in the textile industry
The textile industry is one of the biggest industries for daily use. It employs many workers and has a high production and export volume. The growth of this industry impacts the economy and job opportunities. The quality of fabric products impacts their price. This, in turn, affects the entire industry’s growth. So, textile quality inspection is crucial in the fabric industry chain. AI image recognition and data sharing can enhance fabric quality in textile inspections.
Use AI image recognition to prevent shrinkage in clothes
2. Significance of Fabric Defect Detection
Detecting fabric defects is key in textile inspection. Its main goal is to prevent defects from lowering cloth quality. This, in turn, has a significant effect on the value and sales of textiles.
For a long time, cloth quality monitoring relied on manual visual checks. Staff used their own experience to judge fabric quality, but this method has many flaws. First, the lack of mechanization means that inspecting fabric by hand takes a long time. Second, the evaluation method is not always the same. Inspectors’ biases can cause mistakes and lead to missed defects.
As production ramps up, these issues become clearer. This drives a stronger need for automated fabric inspection solutions.
3. From manual empirical judgment to artificial intelligence recognition
Right now, research on automatic fabric defect detection using AI image recognition is on the rise. Researchers are moving away from manual methods. There are two main types of approaches:
- Traditional image processing techniques for finding defects.
- Deep learning methods for detecting and locating flaws.
Target detection traditionally involves several steps in a systematic process.
- Extract features
- Identify the target.
- Localize the target
Researchers treat feature extraction and target detection as two separate parts.
Deep learning for target detection involves two steps. First, it extracts depth features from images. Then, it localizes targets using convolutional neural networks.
SmartShrink: Shrinkage Rate Tester Use in AI Image Recognition
(To learn more about SmartShrink, click on the video to watch it!)
Fabric surface defects detection and analysis
Normally, every abnormal part of the fabric’s surface is a fabric defect.
Fabric defects usually come from machine faults, yarn issues, or oil stains. Common types include:
- Broken warp and weft defects
- Thick and thin, warp and weft defects
- Breakage defects
- Puckering defects
- Holes defects
- Stain defects
These problems can affect the quality of the fabric.
As fabric patterns grow more intricate, the variety of fabric defects rises. At the same time, advances in textile technology are shrinking the size of these defects. The figure shows some typical fabric defects related to quality standards.
Defect Images on the Surfaces of Various Patterned Fabrics
Textiles go from spinning to finished product through several steps. First, someone spins the yarn. Then, it undergoes cutting, pattern printing, and dyeing. Each of these steps requires many procedures to finish the process. If the conditions aren’t suitable in each construction link, defects can occur. This may happen due to unstandardized staff actions or machine failures. The increase in processes raises the chance of defects.
Common Types of Fabric Defects and Their Causes
| Defect Type | Description | Cause of Formation |
|---|---|---|
| Hole | Holes in the fabric or finished cloth, with broken yarns. | 1. Collision with sharp objects during transport or storage; 2. Damage caused by improper tool use. |
| Oil Stain | Oil stains in strips, spots, or patches on the fabric. | 1. Oil contamination during weaving or printing processes; 2. Oil contamination during transport. |
| Uneven Pilling | Uneven fuzz or pilling, often in irregular shapes or densities. | Sensitive yarns breaking during stretching or uneven tension during processing. |
| Dropped Stitch | Irregular gaps or spaces in the yarn rows, usually larger than normal. Horizontal yarn rows appear broken or misaligned. | Loops are not formed properly during knitting, or yarns are not caught by needles, causing missing loops. |
| Float | Yarns not woven according to the pattern create unbound areas on the surface. | Incorrect loom setup, slack yarn tension, or special float patterns. |
| Knot | Visible knots or thick yarn balls on the fabric surface. | Knots remaining after yarn joining, with ends not trimmed flush. |
| Bar | Noticeable stripe patterns formed by periodic yarn density variation, creating checkerboard or bar effects. | 1. Incorrect yarn feed angle; 2. Uneven tension between the upper and lower warp layers. |
| Slub | Short segments of thick yarn (5–20 mm) on the fabric, larger than normal yarn thickness. | 1. Uneven drafting during spinning; 2. Incomplete slub removal in spinning or winding. |
| Waviness | Ripple-like distortions in the fabric are caused by inconsistent tension during knitting. | Improper take-down tension or ring setting during knitting; damaged or worn-out sinkers or take-down rollers. |
With advances in science and technology, textile fabric technology is improving. As a result, fabric defects are happening less often. But this also makes it trickier to spot any remaining defects. The part of the defect is too small; the previous method is difficult to detect it.
Types of Fabric Defects That Are Difficult to Detect
Image Acquisition and Database Building
Deep learning methods for fabric defect detection are faster and more accurate than traditional ones. They also have a low false detection rate. But these methods need a large amount of training data. In the training phase, use as many images of fabric defects as possible. Include each type of defect in the training network. This helps the model learn about the defects. It will memorize the defect locations and their features. As a result, the model will detect fabric defects with greater speed and accuracy in the future. Faster and more accurate detection of the location of the defects and marking.
First, we set up the fabric image acquisition system. This system includes a light source, lens, camera, image processing card, and actuator. Next, we collect fabric defect images like holes, oil stains, and uneven piling. We then enhance these images using transposition, Gaussian filtering, and image enhancement. This helps us create a fabric image library. It also provides sample support for deep learning later.
Overall Structure Diagram of the Fabric Image Acquisition System
Camera Selection
An industrial camera is vital for the image acquisition system. Good or bad bytes from the camera impact all the following work. Its main goal is to capture a digital image signal. Choosing a camera is crucial. The right camera has a direct impact on image quality. It also impacts how the whole system operates later.
Lens Selection
Lens Selection
The lens is as important as the industrial camera in the image acquisition system. It has a direct impact on the quality of captured images. This, in turn, affects the accuracy of defect detection and processing.
Lenses come in various types. They categorize themselves by their optical standards and application needs. Picking the right lens helps achieve clear, distortion-free images. This is crucial for AI-based fabric inspection systems to work well.
Lighting-Selection
Lighting Selection
Choosing the right illumination is a key part of the image acquisition system. Common choices are fluorescent lamps and LED lights. Several performance factors assess each option.
For fabric defect detection, a stable and long-lasting light source is essential. LED lights usually do better than fluorescent lamps. They offer steadier light, last longer, and use less energy.
| Feature | Fluorescent Lamp | LED Lamp |
|---|---|---|
| Light Efficiency | 50 LM/W | 100 LM/W |
| Flicker Frequency | Tens of Hz (causes image striping) | Constant lighting possible |
| Spectrum | Fixed | Flexible selection |
| Lifespan | Around 10,000 hours | Over 50,000 hours |
Light Source Positioning
After choosing the type of light, the next important step is to position the lights. The position of the light source is key to image quality. It directly impacts the contrast between defective and non-defective areas in the fabric.
To capture images of fabric defects, we use two types of lighting: reflective and transmissive.
- In reflective lighting, the photographer sets the light at an angle above the fabric. The fabric surface reflects the light, and the camera captures it.
- In transmissive lighting, the light sits below the fabric. This setup lets light pass through the material before it hits the camera.
Every lighting method has its perks and leads to different image results. The example setups below highlight these differences.
Comparison of Effects Under Different Light Sources
Database Construction
The TILDA fabric image database has fabric images featuring a mix of background textures. This study used 185 defect images. They all have a uniform plain weave background.
However, the dataset presents some significant challenges. The background looks even, but the fabric texture in good areas is unclear and has uneven spots. Sometimes, even without real defects, the texture and grayscale values can differ from the background. This inconsistency can harm the accuracy of AI defect detection models. It may cause false positives during training or testing.
Partial Defect Images from TILDA Fabric Image Library
Research on Fabric Defect Image Recognition Algorithms
Fabric defect detection is tough. This is a result of the complexity in fabric textures. Traditional algorithms often struggle to achieve both real-time processing and high detection accuracy. Convolutional neural networks (CNNs) have emerged as a great solution to this problem.
Defect Localization Using SSD-Based Neural Networks
To detect and localize with the Single Shot MultiBox Detector (SSD) neural network, do the following:
Step 1:
Split the dataset. Use 80% for training and validation—80% of that is for training and 20% for validation. The last 20% will be the test set to check model performance.
Step 2:
Feed fabric images into the trained detection model. The model detects features and creates bounding boxes that may indicate fabric defects.
Step 3:
Apply a predefined confidence threshold to filter candidate boxes. Next, use the IoU threshold to pick the final bounding boxes that highlight defect areas. The system outputs the coordinates and visualizes the defect locations.
This SSD-based algorithm adapts well. It is very effective in detecting both plain weave and patterned fabrics. It has many uses, high accuracy, and lowers errors from manual checking. The model is also easy to train and deploy.
Detection Results of Fabric Defect ImagesDetection Results of Fabric Defect Images
Conclusion
Deep learning is growing fast, and computing power is getting better. So, CNN-based solutions are being used more often in industries. AI image recognition is changing how we detect fabric surface defects. This technology is improving quality control in textiles. AI image recognition does more than spot visible flaws like holes or stains. It also helps detect fabric shrinkage, which is an important quality parameter. Shrinkage testing is done by measuring fabric before and after washing. This method is labor-intensive and often leads to mistakes. AI systems can capture and analyze changes in fabric images without human intervention. This leads to better accuracy and efficiency. This lets manufacturers track shrinkage rates in real time. They can also ensure that they meet product standards. Smart inspection technologies are evolving. AI image recognition will lead to better textile automation. This will enhance production quality and improve efficiency in the industry.
For more information on textile testing methods/standards or textile testing machines, contact us:
WhatsApp: +86 180 2511 4082
Tel: +86 769 2329 4842
Fax: +86 769 2329 4860
Email: [email protected]