A LLM-Assisted Framework for Hub Casting Character Recognition

Traditional hub casting character recognition heavily relies on manual labor, which is not only labor-intensive and time-consuming but also inefficient. To make the recognition process become automatically, a Large Language Model (LLM)-assisted framework for Hub Casting Character Recognition is proposed. The framework is composed of three modules, LLM-assisted data augmentation module, bolt hole localization module, and OCR module. In the LLM-assisted data augmentation module, text prompts and diffusion model is used to generate diverse and realistic training image samples. In the bolt hole localization module, YOLOv11 is employed for locating bolt holes in the images and outputting the coordinate information of each bolt hole area. In the OCR module, based on the bolt hole area coordinates provided by YOLOv11, the images are further processed to correct potential orientation deviations to ensure the correct alignment of characters, and the orientation-corrected images are input into the PaddleOCR toolkit for character recognition and extraction. The entire framework integrates cutting-edge technologies in the field of artificial intelligence, enabling efficient and accurate recognition of hub casting characters.