Improving waterline control in paper manufacturing using instance segmentation

The waterline is an optical phenomenon observable in the papermaking process. It appears as a visual border between a darker, glossy surface and a lighter one, marking the end of paper structure formation. Analyzing the position and shape of the waterline is useful for operators to optimize paper machine settings, which can impact paper quality. This thesis focuses on developing a computer vision system to provide accurate and accelerated visual inspection via a dashboard accessible to paper machine operators and process engineers for waterline measurement. In this work, different methods, mostly instance segmentation models, are being evaluated and compared. Additionally, the thesis covers data collection procedures, the use of homography for perspective transformation and techniques for increasing model performance, such as data augmentation.