Please use this identifier to cite or link to this item:
Title: Development of an intelligent inspection system for detecting and classifying garment defects
Authors: Zhang, Yihong
Degree: Ph.D.
Issue Date: 2013
Abstract: Nowadays, quality inspection of textile products is an important activity for garment manufactures because the clothing industry is continually pressing for higher product quality and improved productivity to meet both customer demands and reduce the costs associated with quality. Higher production speeds make the timely detection of garment defects more important than ever. So it is necessary to research and develop one new intelligent inspection system for detecting and classifying garment defects. In the present study, a smart garment defect inspection system composed of three components, namely (1) a smart hanger, (2) a shade variation detection unit as well as (3) a stitching-workmanship defect classification unit was developed. Introduction of the intelligent inspection system was proposed in Chapter 1. Literature review work was conducted in Chapter 2. Chapter 3 presented some methodologies about the intelligent inspection system including: (1) To establish System Modeling, (2) To develop one smart hanger model based on 6-DOF robot control,(3) To develop one image stitching model for images capturing by CCD camera, (4) To develop an intelligent detection and classification model for the garment defects. A smart hanger with 6-DOF robot control was developed and introduced in Chapter 4. The proposed robotic hanger has three groups of linkages, namely (1) body link, (2) shoulder link and (3) sleeve link, which are designed to satisfy the conditions of inspection for various garments. To extend the garment more efficiently and more accurately, one optimal fuzzy self-adaptive PID method was conducted in the present study, and the FS-PID controller was applied to implement force control, and the controller performance was validated through simulations and experiments. Apart from the PID controller, the system nonlinearity was compensated by using an adaptive scheme employing the radial basis function (RBF). The stability and tracking result provided in present study was regional in system states. Simulations were given to illustrate the effectiveness of the proposed control scheme. Chapter 5 presented some techniques for image stitching from sequences of images captured by cameras from different angle in garment defects detecting area. The image stitching representation associated a transformation matrix with each input image. In this study, it formulated stitching as a multi-image matching problem, and used invariant local features to find matches between all of the images. The method was insensitive to the ordering, orientation, scale and illumination of the input images. It was also insensitive to noise images that are not part of a panorama, and can recognise multiple panoramas in an unordered image dataset. Detection work of defects played an important role in the automated inspection system for fabrics and garment products. And the detection function is based on image processing by computer. In Chapter 6, three image processing methods for defect detection particularly are applied and introduced, namely (1) the optimal Gabor filters Methods; (2) the Fourier transform method; and (3) optimal color ring projection method. Some method analysis and experiment were also conducted in chapter 6. While in the actual applications, those methods should be improved and optimized according to the different fabric material and different defect types relatively, and some intelligent mathematic model will also be constructed and proposed to match up these images processing method. Good image processing results for defect detection will benefit to the defect classification application work in the next step research work.
Chapter 7 introduced a new intelligent fabric defect detection and classification model based on genetic algorithm (GA) method and Elman neural network was proposed for garment defect detecting. In order to detect and classify garment stitching defects, it was necessary to segment them from the texture background accurately. The textural analysis methods based on the extraction of texture features in the spatial and spectral domains resulted in high dimensionality. Although the methods not relying on textual features were successfully applied to thick fabric defect detection yet, they were not effective in thin surface anomalies as there was a lack of whole structure research and considerations for garment inspection model. A hybrid model using Elman neural network for classifying defects was applied in this section. In Chapter 8, an intelligent fabric defect detection and classification model using genetic algorithms and the modified Single-hidden Layer Feed-forward Neural Network (SLFNN) was introduced. An optimal Gabor filter was proposed for image processing, and the solution for optimization of parameter was based on the niche Genetic Algorithm (NGA) method. The newly modified SLFNN was proposed to classify the type of fabric defects based on a novel algorithm called Regularized Extreme Learning Machine (RELM). The algorithm not only avoids a number of iterations and the local minimum, but also has better generalization, robustness and control-ability than the original ELM. Experiment results showed that the proposed inspecting model was more feasible and applicable in fabric and stitching garment defect detection and classification than the normal back propagation (BP) neural network and ELM neural network. Chapter 9 made the final conclusion for present study and gave some suggestions for future research work. The analyses, findings and contributions of the work completed in previous chapters were summarized; and the limitations and the improvement directions will also be presented in the final Chapter 9.
Subjects: Textile fabrics -- Quality control.
Textile fabrics -- Defects.
Hong Kong Polytechnic University -- Dissertations
Pages: xvii, 208 leaves : ill. ; 30 cm.
Appears in Collections:Thesis

Show full item record

Page views

Last Week
Last month
Citations as of Jun 4, 2023

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.