Integrating dung beetle optimization with deep feature Fusion Model for automated fabric defect Detection and Classification

Quality inspection has revolutionized modern industrial manufacturing. Automated fabric inspection is paramount for maintaining the quality of the fabric in the classical textile industry. In the textile industry production, defect detection often employs visual and manual inspection. But the manual detection is expensive and time-consuming for large-scale production and is prone to subjective judgments. Standard visual techniques are also limited in handling highly variable and non-structured imperfections, and they do not have adaptability and flexibility when facing masses of production data processing. Owing to the emergence of deep learning (DL) and machine vision, automated detection methods are made possible in the industry. However, the major problem of automatic detection methods is to overcome the challenges of missed detection rates and false alarm rates to optimize the stability and accuracy of textile defect detection. The study proposes a novel Dung Beetle Optimization with Deep Feature Fusion for Automated Fabric Defect Detection and Classification (DBODFF-AFDDC) technique. The purpose of the DBODFF-AFDDC technique lies in the effective detection and classification of fabric defects (FDs). To achieve this, the DBODFF-AFDDC approach preprocesses the input fabric images via Wiener filtering (WF)-based noise elimination and Dynamic Histogram Equalization (DHE)-based contrast enhancement. Furthermore, the proposed technique employs a fusion of feature extractors such as EfficientNetV2S and RegNetX002 for learning intrinsic and complex features from the preprocessed images. For FD detection, the Proposed Model technique utilizes the stacked sparse autoencoder (SSAE) model. Moreover, the proposed model technique implements the DBO model for the optimum hyperparameter selection of the SSAE model. Finally, the detected defective regions from the fabric images are segmented by using the NestNet model. The empirical analysis of the DBODFF-AFDDC technique is tested under the fabric image dataset from the Kaggle repository. The comparison study of the DBODFF-AFDDC technique portrayed a superior accuracy value of 99.95% over other DL techniques.