,
Assistant Professor, Department of Information Science and Engineering, Poojya Doddappa Appa College of Engineering, Kalaburagi, India
Associate Professor, Department of Computer Science and Engineering, Sharnbasva University, Kalaburagi, India
The detection of fabric defects through automation is important for ensuring the quality of the textile material because traditional detection methods are not only very time-consuming but also subjective and cannot be applied during the process of production. The traditional models that use techniques of image processing and deep learning have various limitations with respect to the detection of minor texture alterations, structure-based defects, and real-time defect identification in industrial settings. In order to address these issues, an innovative method of fabric defect detection based on the FTEFD model and ResNet classifier is proposed. What is particularly notable about the proposed technique is that it uses not only traditional handcrafted feature descriptors but also residual deep learning approaches. After this, the feature vector is fed into the ResNet classifier using residual blocks, fully connected layers, and softmax classification to detect various defect categories like holes, spots, broken threads, misweave, and normal cloth. In terms of experiment results, it was found that the proposed system gave a model accuracy of 97.48%, precision of 96.92%, recall of 96.35%, F1-Score of 96.63%, and specificity of 97.90%, which is better than CNN, YOLOv5-based, and ResNet systems. In terms of defect-wise analysis, the proposed system also gave average accuracy of 97.66%. The comparative detection results were precision of 0.92, mAP50 of 0.85, mAP50:95 of 0.55, F1-Score of 0.84, and FPS of 129.
This is an open access article distributed under the Creative Commons Attribution Non-Commercial License (CC BY-NC) License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The statements, opinions and data contained in the journal are solely those of the individual authors and contributors and not of the publisher and the editor(s). We stay neutral with regard to jurisdictional claims in published maps and institutional affiliations.
