TY  - JOUR
TI  - Fabric defect detection: a hybrid CNN-LSTM approach using TGANet for improved classification and traceability
AU  - Jerusha, J. Agnes
AU  - Kumar, C. Agees
T2  - Industria Textila
AB  - The fashion industry is incredibly adaptable and strives to adjust to shifting fashion trends. A critical phase in the textile
industry is ensuring that quality standards are met. The identification and categorisation of fabric flaws is an essential
stage in the production process that keeps any defective fabric off the market. Individuals manually detected the fabric’s
surface flaws; however, this is time-consuming and raises issues with human error. The creation of hybrid systems
based on Textile Generative Adversarial Networks (TGANet) is the result of efforts to improve the accuracy of flaw
detection using image processing studies. To solve fabric pattern classification and detection issues in fabric traceability
and management, Convolutional Neural Networks (CNN) and long short-term memory (LSTM) in deep learning are used
to extract texture features. To enhance the feature extractor, we employ TGANet in this study. Using fabric photos to
train the new CNN-LSTM, the texture features are effectively retrieved, and the fabric categories are correctly identified
using classifiers that have been tuned. Multimodal datasets for textile design are used. To verify the model’s robustness
and generalisation, various data are used, such as training with tiny training sets and varying image sizes. The accuracy
and speed of the suggested network model are 94% better than those of the traditional deep learning classification
techniques.
DA  - 2025/06/30/
PY  - 2025
DO  - 10.35530/IT.076.03.2024171
DP  - DOI.org (Crossref)
VL  - 76
IS  - 03
SP  - 387
EP  - 396
SN  - 12225347
ST  - Fabric defect detection
UR  - https://www.revistaindustriatextila.ro/images/2025/3/10%20JERUSHA%20INDUSTRIA%20TEXTILA%20no.3_2025.pdf
Y2  - 2025/07/08/17:40:50
ER  -