1.

Project Title: Automated testing of product defects in textile industry

2. Broad Subject: Machine learning and image processing

3. Project Duration (in months): 24 months

4. Budget (in thousands): 28,000

5. Project Summary

The defects found during the production of fabrics are identified using machine learning and
image processing techniques. The defects in fabrics found during the quality checking after
the dyeing process and the final inspection are focused. So, defects are identified in both the
reactive and proactive stages. The identification of these defects is done mainly using
machine learning and deep learning processes with the help of image processing techniques.
A few sensors are also used to get more details about the fabric and few defects are found this
way to minimize effort and use it in preprocessing the fabric. The few defects that are
focused on during the dyeing processes are colour mismatch defects because of uneven
concentration of colours and using wrong strengths of dyes. The few defects that are focused
on during the final inspection are warping and wefting defects found during warping and
weaving processes, colour patch defects because of uneven dyeing, stains, and dirt defects
due to improper handling of machines, and transport issues. Yarn defects such as thick, thin
or slub projections are also reviewed.

6. Keywords: Fabric defects, machine learning, image processing

7. Objectives

Can detect defects in fabric using machine learning and image processing techniques. Also,
use of sensors for additional feature extraction.

8. Introduction

Textile industry is the platform or industry in which the development, production, processing,
manufacture, and distribution of textile and fabric materials occur. It is one of the most
booming and advanced industries. New developments and automation processes are
incorporated continuously to provide better quality products and improve overall production.
The textile industry mainly has three stages, the yarn production, the fabric production, and
the garment production. Defects occur in all these stages, but this project mainly focuses on
fabric defects.
Fabric is considered the raw material for the garments and quality of the fabric determines the
overall quality of the final garment. Proactive defect detection from the yarn which is the raw
material used for fabric production helps in saving lot of time, money, and labour. It also
helps in the reduction of wastage in raw materials. Reactive defect detection helps in the
rejecting the piece of fabric before the production of garment and hence ensuring the overall
quality of the garments. Because if these defected products are further continued to produce
the garment, they are rated a worser quality than the raw material used in the fabric and hence
gaining no profit for the producers.
9. Definition of the Problems

Quality checking and defect detection are a crucial part of the production of fabrics. The yarn
from the dyeing process to the weaving process till the finishing stages are checked for faults
and defects by various labourers. Hence quality assurance is done throughout production but
all of them are manual processes. The usage of manual labour for every step costs not only
time but money as well. Hence automation can reduce time and cost of the overall
production. This project helps in the automation of this process using machine learning and
image processing techniques helping in easing the process by warning the labourers even
during production time and hence saving a lot of effort and time.

10. Review of status of Research and Development in the subject

10.1 National Status

After a visit to a fabric production industry in Salem, Tamil Nadu to get a better
insight on how textile industries operate in India, it was observed that all the quality checking
and defect detection happened manually. In each stage where quality checking is essential
labourers use their naked eye in inspecting these fabrics. Some large-scale industries use a
fabric inspection machine where the fabric rolls at a particular speed and a fabric inspector
checks it roughly for defects. So, automation of fabric inspection is not widely practised in
India even in large scale industries.

10.2 International Status

Prasanna BandaraThilan Bandara, Tharaka Ranatunga, Vibodha Vimarshana,

Sulochana Sooriyaarachchi, Chathura De Silva from Department of Computer Science and
Engineering, University of Moratuwa, Moratuwa, Sri Lanka Fabric defect detection is a
significant phase of quality control in textile industry. Automating this process is challenging
due to many fabric types and defect types. Defect detection relies on the identification of
regions that are different from the uniform background. Their intention is to develop a system
to detect defects in uniform textured fabrics with Image Processing techniques and Neural
Networks. The results indicated that the use of light beams based on the colour of the
material is more effective than the white light beam.

11. Novelty Importance of the proposed project in the context of current status

Real time defect detection, colour hues detection, yarn quality checking, image processing
techniques
Easy to use, cost and time savings during production, overall quality improvement.

12. Patent details (domestic and international), if applicable

NIL
13. Work plan and Detailed technical information

13.1 Methodology

The collected fabric dataset is initially checked for any corrupted pixels, and it is further pre-
processed. The preprocessing techniques, like image scaling, pixel normalization and
dimension reduction, are used for the training dataset. Feature extraction by the CNN model
aids in characterizing and analysing the defective and non-defective texture of the fabric
images. Datasets consisting of defective and non-defective fabric images are used to train the
system for defect detection and classification, If the fabric image is clean, it is classified as
good else it is defective. The defective images are further classified into colour, cut, hole,
thread, and metal contamination.

Implement image segmentation algorithms to separate the fabric regions from the background
and identify potential defect regions. Apply morphological operations to refine the detected
regions and extract relevant features that characterize defects. Develop methods to extract
both low-level and high-level features from the segmented defect regions, including texture,
colour, shape, and spatial patterns. Utilize techniques like Histogram of Oriented Gradients
(HOG), Local Binary Patterns (LBP), and colour histograms to capture distinctive defect
features. Implement an integrated system that takes fabric images as input, processes them
using image processing techniques, extracts features, and employs the trained machine
learning model to detect and classify defects. Ensure real-time or near-real-time processing,
enabling swift and automated quality assessment during fabric production. Evaluate the
performance of the developed system using appropriate metrics, comparing its accuracy with
manual inspection and other automated methods. Fine-tune the system parameters and
algorithms to achieve optimal results across various fabric types and defect severities. Test
the system's robustness against varying lighting conditions, fabric textures, and defect sizes.
Assess the system's ability to generalize to new, unseen fabric samples, indicating its
practical applicability in real-world scenarios. Thus, image processing algorithms in machine
learning are used and the defects in fabrics are detected and classified.
A few feature extraction processes are also done during the fabric production stage. The raw
material of the fabric that is the yarn is first dyed and then it is weaved to get the desired
fabric. But if the proportion of the dyeing goes a little amiss then the desired tone of shades is
not produced. This is waste of a lot of time, manpower and raw materials. Identifying dyeing
defects in earlier stages also helps in quality inspection of the fabric eventually. A reflective
colour sensor module is used for this purpose which can find the colour composition and hues
using the photoelectric effect. The sensor emits light from a transmitter, and then detects the
light reflected from the detection object with a receiver.
Use of laser sensors for the purpose of measuring the thickness of yarn helps in eradicating
defects related to thick, thin weaving in the middle of the fabric. One of the main benefits of
laser thickness measurement is its accuracy. The use of a laser allows for highly precise
measurements, with some systems capable of measuring thicknesses to within microns.
Another advantage of laser thickness measurement is its speed. Laser thickness measurement
systems use various techniques to measure the distance between the material surface and the
reference point. Some systems use triangulation, where the laser beam is directed at an angle
to the material surface and the angle of reflection is used to calculate the distance. Other
systems use time-of-flight measurements, where the laser pulse is emitted, and the time taken
for the pulse to reflect is used to calculate the distance.

14. Time schedule of activities giving milestones

Processes Months
Requirements 1-2
Design 2-5
Hardware development 5-8
Database Setup 8-10
Image preprocessing and enhancement 10-13
API 13-15
Machine Learning 15-18
Integration 18-20
System Integration Testing 20-22
User Testing 22-23
Deployment 23-24

14.1 Time Schedule of Activities through BAR Diagram

15. Deliverables

 Fully configured hardware

 Machine learning model for defect detection
 Image processing module
 Feature extraction using sensors

16. Target beneficiaries of the proposed work

Textile industries of large scale and heavy workload

17. Suggested plan of action for utilization of research outcome expected from the
project

17.1 As journal publication

The paper will be published once the project is completed.

17.2 Patent filing:

New technology ideas can be patented.

17.3 Project preparation for submission to external funding

The project can be submitted to external funding for real time implementation in textile
industries.
18. References

[1] K. Okamoto and K. Yanai. An automatic calorie estimation system of fruit images on a
smartphone. In Proc. of ACM MM Workshop on Multimedia Assisted Dietary Management,
2016.
[2] https://scantech.com/information/laser-thickness-measurement/
[3] R. Tanno, T. Ege, and K. Yanai. AR DeepCalorieCam V2: Fruit calorie estimation with
cnn and ar-based actual size estimation. In Proc. Of ACM Symposium on Virtual Reality
Software and Technology (VRST), 2018.
[4] https://www.textileindustry.net/different-types-of-fabric-defects/
[5] https://textilelearner.net/weaving-faults-in-fabric/
[6]https://www.researchgate.net/publication/
261422589_An_intelligent_reflective_colour_sensor_system_for_paper_and
_textile_industries
[7] W. Shimoda and K. Yanai. CNN-based fruit image segmentation without pixel-wise
annotation. In Proc. of IAPR International Conference on Image Analysis and Processing,
2015.
[8] https://iopscience.iop.org/article/10.1088/1742-6596/2224/1/012010
[9] R. Tanno, K. Okamoto, and K. Yanai. DeepFruitCam: A DCNNbased real-time mobile
fruit recognition system. In Proc. of ACM MM Workshop on Multimedia Assisted Dietary
Management, 2016.
[10] http://ieeexplore.ieee.org/document/8615491/
[11] C. Rother, V. Kolmogorov, and A. Blake. GrabCut: Interactive foreground extraction
using iterated graph cuts. ACM Trans. on Graphics, 23(3):309–314, 2004.
[12] https://www.mdpi.com/1424-8220/18/4/1064
[13] https://wires.onlinelibrary.wiley.com/doi/abs/10.1002/widm.1228
[14] https://www.emerald.com/insight/content/doi/10.1108/IJCST-11-2018-0135/full/html
[15] https://textilestudycenter.com/fabric-inspection/

19. List of facilities and Equipment available with Department for the project

Computers/ laptops

20. Budget Estimates

S.no. Components Price

1. Cloud storage (5GB per month for 1 year) Rs. 412
2. AutoML vision Rs. 4134
3. Deep learning VM images Rs. 4134
4. Apigee API Rs. 2069
5. Compute engine (c3 standard-4 machine type) Rs. 4961
6. MongoDB Rs. 1653
7. Colour sensor Rs. 900
 8. Laser sensor Rs. 3600
9. Raspberry pi (with camera) Rs. 4500
10. Buzzer and miscellaneous hardware Rs. 1000

21. Budget Justification

Cloud storage – Needed for storing large amounts of data for training the model.

AutoML vision – This enables to train the machine learning models to classify images.

Deep learning VM images – virtual machine images optimized for data science and machine
learning tasks. All images come with key ML frameworks and tools pre-installed and can be
used out of the box on instances with GPUs to accelerate data.

Apigee API – google cloud’s native API management.

Compute engine (c3 standard- 4 machine type) – an infrastructure as a service (IaaS)

to run virtual machines

MongoDB – MongoDB atlas is multi cloud developer data platform. It is used for
managing cloud data base

Colour sensor – to detect defects in colour hues

Laser sensor – noncontact sensor for finding defects in yarn thickness

Raspberry pi (with camera) – for uploading images to cloud (hardware component)