Detection and classification of steel surface defects by image processing and deep learning methods

Metals are elements that have become indispensable in our lives. Metals are used in many areas such as industry, transportation, technology, construction sector, energy sector in daily life. The reason why metals are used so much is that they are durable and long-lasting. However, after a certain period of use on metals, it may become partially or completely unusable. In this case, it reveals the fact that this substance, which is used in almost every field, will cause great problems. In order to prevent possible problems, people have controlled these metals for centuries and intervened where they deem necessary. This process was both time-consuming, expensive and dangerous in terms of perspective that differs from person to person. Scientists dealing with this issue started to produce solutions to the problems that may occur on metal surfaces under the title of visual inspection, with scientific approaches, with a fast, cheap and unchanging point of view. For visual inspection, scientists have applied to image processing methods. Here, scientists used methods such as vortex method and laser diagnosis. Deep learning methods have been put forward with developing methods. In this study, fault classification and fault detection studies were carried out in order to diagnose faults on metal surfaces using deep learning methods. As a result of fine adjustments made on the basis of the VGG-16 model for classification, fault classification has been achieved with high success rates. In the object detection study, different YOLO models were compared against metal surfaces and it was determined that the model that gave the most accurate and fastest results was the YOLO-V5.