IRJET- Fabric Defect Classification using Modular Neural Network

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 04 | Apr 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 3353
FABRIC DEFECT CLASSIFICATION USING MODULAR NEURAL NETWORK
Miss.Rupali N.Tirale1, Dr.V.L Agrawal2,Prof.Y.P Sushir3
1Student, Electronic and Telecommunication of Padm. Dr. V.B Kolte College of Engineering, Malkapur (India)
2Professor,Electronic and Telecommunication of HVPM’S College of Engineering and Technology (India)
3Assistant Professor, Electronic and Telecommunication of Padm. Dr. V.B Kolte College of Engineering,
Malkapur (India)
----------------------------------------------------------------------***---------------------------------------------------------------------
Abstract: In this paper a new classification algorithm is
proposed for the Fabric Defect. In order to develop
algorithm 164 different fabric images With a view to
extract features from the imagesafter image processing, an
algorithm proposes (WHT)Wavelet Transform coefficients.
The Efficient classifiers based on Modular neural
Network (MNN). A separate Cross-Validation dataset is
used for proper evaluation of the proposed classification
algorithm with respect to important performancemeasures,
such as MSE and classification accuracy. The Average
ClassificationAccuracy of MNN Neural Networkcomprising
of one hidden layers1 with 8 PE’s organized in a typical
topology is found to be superior (92.65 %) for Training.
Finally, optimal algorithm has been developed on the basis
of the best classifier performance. The algorithm will
provide an effective alternative to traditional method of
fabric defect analysis for deciding the best quality fabric.
Keywords— MatLab, Neuro Solution Software, Microsoft
excel, WHT Transform Techniques
1. INTRODUCTION
In the manufacturing process, if the cost and just-in-time
delivery represent the two lines of the right angle, the
quality should be the hypotenuse that completes the right
triangle of the process. It means that the quality is the most
important parameter despite the increase in one or both of
the other parameters (geometrical fact). Scientifically, a
process qualitycontrol meansconductingobservations,tests
and inspections and thereby making decisions which
improve its performance. Because no production or
manufacturing process is 100% defect-free (this applies
particularly where natural materials, as textile ones, are
processed), the success of a weaving mill is significantly
highlighted by its success in reducing fabric defects.
For a weaving plant, in these harsh economic times, first
quality fabric plays the main role to insure survival in a
competitive marketplace. This puts sophisticated stress on
the weaving industry to work towardsa lowcostfirstquality
product as well as just-in-time delivery.First qualityfabricis
totally free of major defects and virtually free of minor
structural or surface defects. Second quality fabric is fabric
that may contain a few major defects and/or several minor
structural or surface defects [1]. The non-detected fabric
defects are responsible for at least 50% ofthesecondquality
in the garment industry (this figure is the result of many
years of practical experience), which represents a loss in
revenue for the manufacturers since the product will sell for
only 45%-65% the price of first Quality product, whileusing
the same amount of production resources.
Although quality levels have been greatly improved withthe
continuous improvement of materials and technologies,
most weavers still find it necessary to perform 100%
inspection because customer expectations have also
increased and the risk of delivering inferior quality fabrics
without inspection is not acceptable. The key issue,
therefore, is how and under what conditions fabric
inspection will lead to quality improvement. To address this
issue, we proposed this classification system.
The modern weaving Industry faces a lot of difficult
challenges to create a high productivity as well as high-
quality-manufacturing environment. Because production
speeds are faster than ever and because of the increase in
roll sizes, manufacturers must be able to identify defects,
locate their sources, and make the necessary corrections in
less time so as to reduce the amount of second qualityfabric.
This in turn places a greater strain on the inspection
departments of the manufacturers. Due to factors such as
tiredness, boredom and, inattentiveness, the staff
performance is often unreliable. The inspector can hardly
determine the level of faults that is acceptable, but
comparing such a level between several inspectorsisalmost
impossible. Therefore, the best possibility of objective and
consistent evaluation is through the application of an
automatic inspection system.
From the early beginning, the human dream is to improve
the manufacturing techniques to achieve optimumpotential
benefits as quality, cost, comfort, accuracy, precision and
speed. To imitate the wide variety of human functions,
technology wasthemagicstick thatadvancedhumanityfrom
manual to mechanical and then from mechanical to
automatic. The rare existence ofautomatedfabricinspection
may be attributed to the methodologies, which are often
unable to cope with a wide variety of fabrics and defects, yet
a continued reduction in processorandmemorycostswould
suggest that automated fabric inspection has potential as a
cost effective alternative. The wider application of
automated fabric inspection wouldseemtooffera numberof
potential advantages, including improved safety, reduced

labor costs, the elimination of human error and/or
subjective judgment, and the creation of timely statistical
product data. Therefore, automated visual inspection is
gaining increasing importance in weaving industry.
An automated inspection system usually consists of a
computer-based vision system. Because they are computer-
based, these systems do not suffer the drawbacks of human
visual inspection. Automated systems are able to inspect
fabric in a continuous manner without pause. Most of these
automated systems are offline or off-loom systems. Should
any defects be found that are mechanical in nature (i.e.,
missing ends or oil spots), the lag time that exists between
actual production and inspection translates into more
defective fabric produced on the machine that is causing
these defects. Therefore, to be more efficient, inspection
systems must be implemented online or on-loom.
The Proposed method in this synopsis represents an
effective and accurate approach to automatic defect
detection. It is capable of identifying all five type defects.
Because the defect-free fabric has a periodic regular
structure, the occurrence of a defect in the fabric breaks the
regular structure. Therefore, the fabric defects can be
detected by monitoring fabric structure. Fourier Transform
gives the possibility to monitoranddescribetherelationship
between the regular structure of the fabric in the spatial
domain and its Fourier spectrum in the frequency domain.
Presence of a defect over the periodical structure of woven
fabric causes changes in its Fourier spectrum. By comparing
the power spectrum of an image containing a defect with
that of a defect-free image, changes in the normalized
intensity between one spectrum and the other means the
presence of a defect.
The fabric defect could be simply defined as a change in or
on the fabric construction. Only the weaving process may
create a huge number of defects named as weaving defects.
Most of these defects appear in the longitudinal direction of
the fabric (the warp direction) or inthewidth-wisedirection
(the weft direction). The yarnrepresentsthe mostimportant
reason of these defects, where presence or absence of the
yarn causes some defects such as miss-ends or picks, end
outs, and broken end or picks. Other defects are due to yarn
defects such as slubs, contaminations or waste, becoming
trapped in the fabric structure during weaving process.
Additional defects are mostlymachinerelated,andappearas
structural failures (tears or holes) or machine residue (oil
spots or dirt). Because of the wide variety of defects as
mentioned previously, it will be gainful to applythestudy on
the most major fabric defects. The chosen major defects are
hole, oil stain, float, coarse-end, coarse-pick, double-end,
double-pick, irregular weft density, broken end, and broken
pick.
1.1 Defect Analysis
In this proposed work, we have dealt with four types of
defect, which often occur in knitted fabrics in Bangladesh,
namely color yarn, hole, missing yarn, and spot. All of the
defects are shown in Fig. 1. All of them are discussed here
below.
(a) (b)
(c) (d)
Figure 1: Different types of defect occurred in knitted
fabrics.
(a) Bunching Up. (b) Hole. (c) Missing yarn. (d) Spot.
• Bunching Up: Fig. 1(a) shows the defect of Visible knots in
the fabric are referred to as bunching up. They appear as
beads and turn up irregularly in the fabric. Can build up
resulting in a ‘cloudy’ appearance. More irregular the yarn,
more pronounced is the ‘cloudy’ appearance.
• Hole: Fig. 1(b) shows the defect of hole. Hole appears in a
shape, close to a circle of the color of the background, on a
fabric of another color. Its size varies from small to medium.
Background color is another issue. In some cases,
background color can become close to fabric color.
• Missing Yarn: Fig. 1(c) shows the defect of missing yarn.
Missing yarn appears as a thin striped shade of the color of
fabric. It is usually long. It is of two types, namely vertical
and horizontal
• Spot: Fig. 1(d) shows the defect of spot. Spot does not
appear in any specific shape. It usuallyappearsina scattered
form of one color on a fabric of another color. Moreover, its
size varies widely. A camera of high resolution and proper
lighting are required in order to clearly capture the image of
the defect of spot.

II. RESEARCH METHODOLOGY
Figure2.1: Methodology of work
It is characterization of four type of fabric defect images
Using Neural Network Approaches.. Information
procurement for the proposed classifier intended for the
order of fabric defect images. The most essential unrelated
highlights and in addition coefficient from theimageswill be
removed .keeping in mind theendgoal toseparatehighlights
WHT transform will be utilized.
2.1Neural Networks
Following Neural Networks are tested:
Modular Neural Network (MNN)
Modular Neural Network is in fact a modular feed forward
neural network which is a special category of MLP NN. It
does not have full interconnectivity between their layers.
Therefore, a smaller number of connection weights may be
required for the same size MLP network with regard to the
same number of processing elements. In view of these facts,
the training time is accelerated. There have beenmany ways
in order to segment a MNN into different modules. MNN
processes its inputs with the help of numerous parallel
connected MLPs and the outputs of these MLP are
recombined to produce the results. This neural network is
comprised of different sub modules and according to a
specific topology; some structure is created within the
topology in order to boost specialization of function in each
sub-module.
The following topology depicted in Fig.2.2 of the MNN has
produced the best classification results.
Fig. 2.2: Topology of a Modular Neural Network
This topology is recommended on the basis of experimental
evidences, testing and performance measures.
 Learning Rules used:
Momentum
Momentum simply adds a fraction m of the previous weight
update to the currentone.Themomentumparameterisused
to prevent the system from converging to a local minimum
or saddle point. A high momentum parameter can also help
to increase the speed of convergence of the system.
However, setting the momentum parameter too high can
create a risk of overshooting the minimum, which can cause
the system to becomeunstable.Amomentumcoefficientthat
is too low cannot reliably avoid local minima, and can also
slow down the training of the system.
Conjugate Gradient
CG is the most popular iterative method for solving large
systems of linear equations. CG is effective forsystemsof the
form A= xb-A (1) where x _is an unknown vector, b is a
known vector, and A _is a known, square, symmetric,
positive-definite(orpositive-indefinite)matrix.(Don’tworry
if you’ve forgotten what “positive-definite” means; we shall
review it.) These systems arise in many important settings,
such as finite difference and finite element methods for
solving partial differential equations, structural analysis,
circuit analysis, and math homework.
Developed by Widrow and Hoff, the delta rule,alsocalledthe
Least Mean Square (LMS) method, is one of the most
commonly used learning rules. For a given input vector, the
output vector is compared to the correct answer. If the
difference is zero, no learning takes place; otherwise, the
weights are adjusted to reduce this difference.Thechangein
weight from ui to uj is given by: dwij = r* ai * ej, where r is
the learning rate, ai represents the activation of ui and ej is
the difference between the expected output and the actual
output of uj. If the set of input patterns form a linearly
independent set then arbitrary associations can be learned
using the delta rule.
It has been shown that for networks with linear activation
functions and with no hidden units (hidden units are found
in networks with more than two layers), the error squared
U
1
U1 U2
L1 L2

vs. the weight graph is a paraboloid in n-space. Since the
proportionality constant is negative, the graph of such a
function is concave upward and has a minimum value. The
vertex of this paraboloid represents the point where the
error is minimized. The weight vector corresponding to this
point is then the ideal weight vector.
Quick propagation
Quick propagation (Quickprop) [1] is one of the most
effective and widely used adaptive learning rules. There is
only one global parameter making a significant contribution
to the result, the e-parameter. Quick-propagation uses a set
of heuristics to optimise Back-propagation, the condition
where e is used is when the sign for the current slope and
previous slope for the weight is the same.
Delta by Delta
Developed by Widrow and Hoff, the delta rule,alsocalledthe
Least Mean Square (LMS) method, is one of the most
commonly used learning rules. For a given input vector, the
output vector is compared to the correct answer. If the
difference is zero, no learning takes place; otherwise, the
weights are adjusted to reduce this difference.Thechangein
weight from ui to uj is given by: dwij = r* ai * ej, where r is
the learning rate, ai represents the activation of ui and ej is
the difference between the expected output and the actual
output of uj. If the set of input patterns form a linearly
independent set then arbitrary associations can be learned
using the delta rule.
It has been shown that for networks with linear activation
functions and with no hidden units (hidden units are found
in networks with more than two layers), the error squared
vs. the weight graph is a paraboloid in n-space. Since the
proportionality constant is negative, the graph of such a
function is concave upward and has a minimum value. The
vertex of this paraboloid represents the point where the
error is minimized. The weight vector corresponding to this
point is then the ideal weight vector. [10]
III. SIMULATION RESULTS
1) Computer Simulation
The MNN neural system has been simulated for 164 distinct
images of four type of fabric defect images out of which 148
were utilized for training reason and 16 were utilized for
cross validation.
The simulation of best classifier along with the confusion
matrix is shown below:
Figure.3.1: MNN1 neural network trained with MOM
learning rule
2) Results
Table I: Confusion matrix on CV data set
Output /
Desired
NAME
(HOLE)
NAME
(SPOT)
NAME
(MISSING
YARN)
NAME
(BUNCHIN
G UP)
NAME
(HOLE) 3 0 0 0
NAME
(SPOT) 0 5 0 0
NAME
(MISSING
YARN) 1 0 4 1
NAME
(BUNCHIN
G UP) 0 0 0 3
TABLE II: Confusion matrix on Training data set
Here Table I and Table II Contend the C.V as well as Training
data set.
Output /
Desired
NAME
(HOLE)
NAME
(SPOT)
NAME
(MISSING
YARN)
NAME
(BUNCHIN
G UP)
NAME
(HOLE) 3 0 0 0
NAME
(SPOT) 0 5 0 0
NAME
(MISSING
YARN) 1 0 4 1
NAME
(BUNCHIG
UP) 0 0 0 3

TABLE III: Accuracy of the network on CV data set
Perform
ance
NAME(H
OLE)
NAME(S
POT)
NAME(MI
SSING
YARN)
NAME(BUN
CHING UP)
MSE
0.06458
4683
0.00341
3721
0.066869
32
0.0709700
96
NMSE
0.35894
1795
0.01644
2757
0.371639
104
0.3944299
54
MAE
0.15142
9307
0.05753
4789
0.159487
752
0.1334338
28
Min Abs
Error
0.00202
568
0.04664
9977
0.004749
918
0.0278359
63
Max Abs
Error
0.72453
9422
0.08586
516
0.813470
391
1.0406134
32
R
0.92822
6116
0.99583
889
0.809290
479
0.7849967
97
Percent
Correct 75 100 100 75
TABLE IV: Accuracy of the network on training data set
Here Table III and Table IV Contain the C.V and Training
result and show the 92.65% percent accuracy.
IV. CONCLUSION AND FUTURE WORK
From the results obtained it concludes that the MNN Neural
Network with MOM (momentum) and hidden layer 1 with
processing element 8 gives best results of 97.81% in
Training while in Cross Validation it gives87.05%sooverall
result is 92.65%.
V. ACKNOWLEDGMENT
We are very grateful to our Padm. Dr. V.B Kolte College of
Engineering, Malkapur to support and other faculty and
associates of ENTC department who are directly&indirectly
helped me for these papers.
References
[1]J. Tou, R. Gonzalez, Pattern Recognition Principles.
Massachusetts, USA: Addison-Wesley Publishing Company,
1981.
[2] R. Stojanovic, P. Mitropulos, C. Koulamas, Y.Karayiannis,
S. Koubias, and G. Papadopoulos,“Real-time vision based
system for textile fabric inspection,” Real-Time Imaging,vol.
7, no. 6, pp. 507-518, 2001.
[3] R. Saeidi, M. Latifi, S. Najar, and A. Saeidi, “Computer
vision-aided fabric inspectionsystemforon-circularknitting
machine,” Textile Research Journal, vol. 75, no. 6, pp. 492-
497, June 2005.
[4] A. Islam, S. Akhter, and T. Mursalin, “Automated textile
defect recognition system using computer vision and
artificial neural networks,” Proceedings World Academy of
Science, Engineering and Technology, vol. 13, pp. 1-7, Jan.
2006.
[5] V. Murino, M. Bicego, and I. Rossi, “Statistical
classification of raw textile defects,” in Proceedings of the
17th International Conference on Pattern Recognition
(ICPR'04), pp. 311-314, 2004.
[6] Y. Karayiannis, R. Stojanovic,P.Mitropoulos,C.Koulamas,
T. Stouraitis, S. Koubias, and G. Papadopoulos, “Defect
detection and classification on web textile fabric using
multiresolution decomposition and neural networks,” in
Proceedings of the 6th IEEE International Conference on
Electronics, Circuits and Systems, pp. 765-768, Sept. 1999,
Cyprus.
[7] A. Kumar, “Neural network based detection of local
textile defects,” Pattern Recognition, vol. 36, pp. 1645-1659,
2003.
[8] D. Karras, S. Karkanis, and B. Mertzios,“Supervised and
unsupervised neural network methods applied to textile
quality control based on improved wavelet feature
extraction techniques,” International Journal on Computer
Mathematics,vol. 67, pp. 169-181, 1998.
[9] C. Kuo, C. Lee, “A back-propagation neural network for
recognizing fabric defects,” Textile Research Journal,vol.73,
pp. 147-151, 2003.
[10] P. Mitropoulos, C. Koulamas, R. Stojanovic, S.Koubias, G.
Papadopoulos, & G. Karayanis, “Real-time vision system for
defect detection and neural classification of web textile

fabric,” in Proceedings of SPIE Conference, vol. 3652, pp.59-
69, Jan. 1999, USA.
[11] E. Shady, Y. Gowayed, M. Abouiiana, S. Youssef, and C.
Pastore, “Detection and classification of defects in knitted
fabric structures,” Textile Research Journal, vol. 76, pp. 295-
300, 2006.
[12] J. Campbell, C. Fraley, D. Stanford, F. Murtagh,and A.
Raftery, “Model-based methods for textile fault detection,”
International Journal of Imaging Systems and Technology,
vol. 10, pp. 339-346, Jul. 1999.
[13] F. Cohen, Z. Fan, and Z. Attali, “Automated inspection of
textile fabrics using textural models,” IEEE Transactions on
Pattern Analysis and Machine Intelligence, vol. 8, pp. 803-
808, Aug. 1991.
[14] J. Campbell, A. Hashim, T. McGinnity, and T.Lunney,
“Flaw detection in woven textiles by neural network,” in
Proceedings of the 5th Irish Neural Networks Conference,
pp. 92-99, Sept. 1995, Ireland.
[15] K. Mak, P. Peng, and H. Lau, “A real-time computer
vision system for detecting defects in textile fabrics,” in
Proceedings of the IEEE International Conference on
Industrial Technology (ICIT'05), pp. 469-474, 14-17 Dec.
2005, Hong Kong, China.
[16] M. Salahudin, M. Rokonuzzaman, “Adaptive
segmentation of knit fabric images for automated defect
detection in semi-structured environments,” in Proceedings
of the 8th International Conference on Computer and
Information Technology, pp.255-260, 2005, Bangladesh.
[17] Y. Shu, Z. Tan, “Fabric defects automatic detectionusing
gabor filters,” in Proceedings of the 5th World Congress on
Intelligent Control and Automation, pp. 3378-3380, 15-19
June 2004, China.
[18] M. Islam, S. Akhter, T. Mursalin, and M. Amin,“A suitable
neural network to detect textile defects,” Neural Information
Processing,SpringerLink, vol. 4233, pp. 430-438, Oct. 2006.
[19] A. Abouelela, H. Abbas, H. Eldeeb, A. Wahdan, and S.
Nassar, “Automated vision system for localizing structural
defects in textile fabrics,” Pattern Recognition Letters, vol.
26, pp. 1435- 1443, July 2005.
[20] W. Jasper, J. Joines, and J. Brenzovich, “Fabric defect
detection using a genetic algorithm tuned wavelet filter,”
Journal of the Textile Institute, vol. 96, pp. 43-54, Jan. 2005.

IRJET- Fabric Defect Classification using Modular Neural Network

More Related Content

What's hot (20)

Similar to IRJET- Fabric Defect Classification using Modular Neural Network (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Fabric Defect Classification using Modular Neural Network