Classification of Surface Defects on Steel Strip Images using Convolution Neural Network and Support Vector Machine

Quality control of the surfaces of rolled products has received wide attention due to the crucial role that these products play in the manufacture of various car bodies, planes, ships, and trains. The process of quality control has undergone remarkable development. Previously, it was based on the human eye and characterized by slowness, fatigue, and error. To overcome these problems, nowadays the quality control is based mainly on computer vision. In this context, we propose in this work to develop an intelligent recognition system of surface defects for hot-rolled steel strips images using modified AlexNet convolution neural network and support vector machine model. Furthermore, we conducted a study on the effect of layers selection on classification accuracy. We have trained and tested our classification model using a public database of Northeastern University composed of 1800 images of defects. The results showed that our classifier model can be used easily for effective screening of surface defects for hot-rolled steel strips with very a high classification accuracy up to 99.7%, using only 7% of the total extracted features for each image with activations on the fully connected layer "FC7." In addition, we addressed through this research a comparative study between the proposed classification model and the well-known modern classification models. This study highlighted the efficiency and effectiveness of our proposed model for the classification of surface defects.