Surface defect detection using deep learning

Defects in steel raise various problems when operating on it and leads to unmitigated disasters. To tackle the issue deep learning is being used and the different types of defects are recognized beforehand for risk management. While manually it is tough for anyone to recognize these disasters, deep learning plays its part by using a pretrained model to extract the features and classify the defects categorically. Deep learning has been making great strides nowadays. Currently deep learning is being used across various research centers and universities. In this paper with the help of models like LeNet, AlexNet and VGG-16. These models help us recognizing the defects in the increasing order of accuracy. Accuracy of up to 93.4% has been attained by VGG 16 models which can be increased by increasing the number of epochs. Shortcomings like overfitting because of the lack of data has been overcome with the help of data augmentation, batch normalization and dropout. Adding this to already existing technology can help us compute the type of defects and what necessary actions we have to take in order to do damage control.