Defective and nondefective classif ication of fabric images using shallow and deep networks

Abstract

The defect detection is an important activity in quality analysis and control in the fabric industry. The presented work gives a comparative analysis of artificial neural network and deep learning architectures. The MobileNet and deep residual network (ResNet) are deployed to classify the defective and nondefective fabric images. The hand-crafted morphological features are used in fabric image analysis along with feed backward selection feature reduction method to obtain the significant features. The overall classification rates of 95.3%, 98.2%, and 99.65% are obtained for Shallow, ResNet, and MobileNet architectures, respectively. The MobileNet model has given a maximum classification rate than Shallow and ResNet architectures. The work finds applications in apparel industry, quality analysis, cost estimation, online purchase of fabric, Industry 4.0, and so on.