Steel Surface Defects Classification Based on Multi-Derivatives Locally Encoded Transform Feature Histogram (LETRIST) Features

Surface quality is a critical factor in the assessment of steel products. However, traditional manual inspection methods often fail to ensure the flawless condition of steel surfaces. Over the past two decades, vision-based techniques for automated steel surface inspection have emerged as powerful solutions to meet the stringent requirements of the industrial sector. In this context, this paper explores the application of the LETRIST (Locally Encoded Transform Feature Histogram) texture descriptor for accurate classification of steel surface defects. Additionally, an enhancement based on a multi-derivative framework is proposed to improve its efficiency. The proposed method involves several steps. Firstly, directional derivative masks are applied to the defect images. Then, LETRIST-based features are computed to represent each resulting image. Principal Component Analysis (PCA) is subsequently applied to these features to reduce the size of the descriptor. Finally, two fusion schemes, namely feature level fusion and decision level fusion, are utilized to determine the true class of the image defect. A series of experiments is conducted on the NEU Surface Defects database to evaluate the performance of the proposed system. Comparative analysis with state-of-the-art steel surface defect classification systems demonstrates that the suggested approach outperforms existing methods in terms of classification accuracy.