In textile commodities, the flaws recognition is the most time-consuming process which discovers the Fabric Defects (FDs) to boost the fabric quality. To combat this issue, an Enhanced Pair wise-Potential Activation Layer in Optimized Convolutional Neural Network (EPPAL-OCNN) model has been recently designed which solves the undesired convergence of CNN to increase the recognition rate. But, OCNN training needs more labeled samples and takes more time to label the fabric samples. Hence, this article designs an EPPAL- Optimized Multi-Criteria CNN (EPPAL-OMCCNN) model based on the multi-objective active sampling mechanism. The main aim of this model is to reduce the labeling time while considering more fabric samples for OCNN training. Initially, the OCNN structure is created using a limited amount of samples. After that, more influential samples are labeled based on the multi-objective sampling mechanism. By using these labeled samples, the OCNN is upgraded to recognize and categorize the FDs with the highest precision.Further, the test samples from the TILDA corpus are used to validate the EPPAL-OMCCNN which reveals that it attains 96.27% of accuracy compared to the classical models.

Fiber defects recognition, EPPAL-OCNN, Image labelling, Active learning, Multi-objective sampling