ScholarWorks@경상국립대학교

초록

This study proposes an image segmentation technique utilizing the U-Net model to effectively segment the cross-sections of corroded specimens. The proposed model, leveraging an encoder-decoder architecture with skip connections, enables high-resolution segmentation, which is advantageous for the precise delineation of complex corroded areas. After training the model on a labeled image dataset, performance evaluation using test images demonstrated that the proposed U-Net model achieved high accuracy and IoU scores, thereby confirming its excellent performance. These results indicate that machine learning-based long-term image analysis can contribute to the efficient and straightforward segmentation of specimens.

키워드