Development of a method of pre-processing digital images to improve the quality of detection of sheet metal surface defects using a neural network model

The issues of detection and classification of surface defects of rolled steel sheets using deep learning and computer vision methods are considered in the paper. Qualitative detection and classification of defects play a key role in improving of quality production standards and attestation of rolled metal products. This paper presents a detection model based on the Faster R-CNN convolutional neural network. To enhance the performance of this model, a specialized image preprocessing algorithm has been developed. The developed algorithm provides high sensitivity of the video inspection system, allowing to detect defects up to 0.5 x 0.5 mm in size, which makes it optimal for application in real-time systems.