Machine Learning Driven Exploration and Identification of Steel Surfaces: Leveraging Advanced Computer Vision Techniques

Steel surface flaw detection is a critical challenge for industries worldwide that have adopted manufacturing quality management. Conventional inspection methods are time-consuming, prompting the exploration of image analysis and machine learning techniques for efficient fault detection. This study aims to develop models for detecting steel defects using the SEVERSTAL dataset, which comprises 12,568 images of steel surfaces. Models were initially created for binary categorization (defect and no defect) and subsequently evaluated for multi-label classification. Additionally, the U-Net architecture was utilized to improve the accuracy of defect detection. Test-time augmentation was applied to further enhance the models' performance. The results demonstrate the effectiveness of the proposed approach in accurately detecting steel defects, highlighting the potential of machine learning and deep learning techniques in improving fault detection processes in the steel industry.