Fabric defect detection is an important step in the fabric production process. Fabric defect detection can be performed visually by users or intelligently. Traditional intelligent methods for fabric defect detection were based on manual feature extraction methods and the use of old classifiers. However, these methods were not accurate enough. With the development of deep learning models, these methods have become very efficient. Various researches have been conducted on the use of DL-based methods for fabric defect detection, but these methods usually do not consider the texture of fabric defects and hence are effective for a specific category of fabric defects. However, fabric defects can be coarse or fine, and the need for a comprehensive network that has the ability to detect both categories of defects is felt in the research literature. In this research, a fabric defect detection method based on two-level convolutional neural networks and Canonical Correlation Analysis is presented. The aim of the proposed method is to fill the previous gaps in research. In previous methods, the size of fabric defects and their texture type were not taken into account. In the proposed method, two neural networks based on attention mechanisms are designed, each of which is customized for a specific category of defects. The results of these two networks are fused using Canonical Correlation Analysis (CCA) to create a single architecture. Simulation results show that the proposed method has an accuracy of 91.35% for detecting fabric defects of any texture type and size.
