A Segmentation Method for Penicillin Bottle Bottom Defects based on U-Net++ and Regional Growth

This paper studies and improves a penicillin bottle bottom defect segmentation method. The method used U-Net++ and region growth method to improve detection accuracy and solve problems, such as slow manual detection speed and poor accuracy. The proposed method first used U-Net++ for image segmentation, and then used region growth algorithm to finely adjust segmentation results. The region growth algorithm automatically selected initial seed points and improved growth criteria, and further optimizes segmentation accuracy using adaptive threshold growth and gradient additional growth conditions. The experimental results showed that the improved segmentation method outperforms traditional segmentation networks in indicators such as average intersection to union ratio (83.8902%), average Hausdorff distance (1.969067), and Dice coefficient (88.9876%). The method significantly improves segmentation performance and detection accuracy.