열간압연 제품의 폭 방향 온도 불량 분류를 위한 준지도 학습

In this study, we propose a semi-supervised learning approach for classifying transverse temperature defects during the hot rolling process. Thermal imaging cameras capture and store thermal images of each product for measuring the transverse temperature variations, which can cause defects in steel pipe manufacturing. To mitigate this risk, thermal defect images are manually classified by individuals. However, classifications by human rely heavily on experience, leading to differences among classifiers. Thus, there is a need for a model capable of automatically classifying transverse temperature images in the hot rolling process. Labeling these thermal images for model training pose a challenge because of the time-intensive nature. To address this challenge and ensure good performance, we propose using a semi-supervised learning approach that leverages both labeled and unlabeled images for actual steel manufacturing data. To the best of our knowledge, this study is the first to apply semi-supervised learning to classify thermal defect images of steel products. The proposed model achieves outstanding performance with an F1-Score of 0.9608 even when trained using only 50% of the labeled data. We demonstrate the effectiveness and practicality of the propose method by using actual thermal images of steel products. We believe that the outcome of this study can lead to significant time and resource savings in defect detection within steel manufacturing processes.