An Optimized YOLOv11-Based Deep Learning Framework With CNN Feature Enhancement for Tile Crack Detection

The tile business is a critical part of the national economic development, as it is one of the areas that provides employment, produces goods, and exports them. Regardless of its significance, there are challenges in the industry that are caused by issues with production, which in most cases is caused by low-quality materials used or by mishandling of the products during transportation. Historically, visitors to the site have been able to detect tile cracks using human eyes, which, in addition to being expensive and time-consuming, can also be unreliable. This paper proposes a trustworthy approach to identifying tile cracks using the YOLOv11 model to solve these challenges. It is a combination of enhanced CNN preprocessing and the YOLOv11 model to enhance the effectiveness of crack detection. It takes advantage of the capabilities of the YOLOv11 model to detect cracks on tiles and aims to use a wide range of tile images in different lighting conditions, textures, and types of defects. The approach uses a powerful tile image analysis approach, and the high detection precision with the bounding box method is 88.30%, and the mask method precision is 88.77% with a small number of false positives.