Explainable ai and optimized feature fusion network for enhanced surface defect detection in steel a deep learning approach

Abstract

newline Steel surface defect recognition has a crucial role in industrial newlineproducts. Surface defect detection has gained increasing significance in recent newlineyears. Over time, the evolution of steel surface defect recognition technology newlinehas progressed from manual detection to automatic detection using traditional newlinemachine learning algorithms. The accurate detection of defects on steel newlinesurfaces is essential to ensure the high quality of steel products. It has been newlinecaused by several aspects: (1) the high production speed improves the realtime newlinerequirements of detection; (2) the surface defects of steel with irregular newlineshapes and numerous small defects increase the difficulty of accurate newlinedetection; (3) Steel strip surface defects come in a extensive variety of shapes, newlinesizes, and degrees of complexity, and flaws resulting from different newlinemanufacturing processes frequently have distinct characters; (4) Segmentation newlineof crack area from non-crack area has been a difficult task due to newlineirregularities, noisy, and discontinuities in the surface; (5) Object detection, newlinesurface defect detection must detect small defects and defects with different newlineaspect ratios; (6) Fine features and the inordinate positional accuracy of the newlinesurface defects make the identification tasks very difficult. Thus newlinesegmentation, feature extraction, classification, and detection algorithms have newlinebeen introduced for the surface defects of steel strips to increase the accuracy newlineand reduce the computation time. newlineThe first contribution of the work, A transfer learning-based algorithm newlineis introduced for detecting cracks on concrete surfaces