Robust and Rapid Fabric Defect Detection Using EGNet

Abstract

The quality of the fabric item is considered as one of the most significant concentrations in the textile industry. Convolutional Neural Network has shown commendable execution of fabric defect identification with computer vision and image handling. In this paper, we have implemented a Reformed Convolution Neural Network architecture known as ‘EGNet’ for fabric defect detection. The model has trained on the Cotton Incorporated dataset with 70% data as training and 30% as validation dataset. The model consists of 22 layers of Convolutional layer and Pooling Layer one after the other. The recognition of fabric faults using EGNet is executed utilizing load image dataset, load EGNet, replace final layers, network training, classify validation images. The EGNet is optimized using stochastic gradient descent with momentum. Data augmentation and max-pooling techniques are used to reduce the network's overfitting issue. To infer the significance of EGNet, comparative analysis is done with AlexNet and the result shows that EGNet architecture exposes the fabric defects in 23 seconds of elapsed time and with 100 percent of accuracy.