Study on the Identification of Wavy Features of Si3N4 Wafer Microdamage With Improved K‐Means Clustering by Incorporating the Centroid Growth Mechanism

To address the problems of irregular wavy edge contours, blurred edges, and notch misidentification in microscratch damage on Si3N4 wafers, an improved K‐means clustering segmentation method with centroid growth fusion is proposed. This method aims to achieve complete segmentation of the wavy edge features of microscratch damage on Si3N4 wafers. Through in‐depth analysis of the intermittent notch characteristics of the wavy edge features of Si3N4 wafers, the initial clustering center is set based on the Euclidean distance probability model and combined with the image hierarchical clustering framework to achieve effective separation of image content. An image feature connectivity assessment mechanism is also introduced to enhance the extraction accuracy of target features. The experimental results show that the precision of the proposed method reaches 97.76% and the MAE reaches 3.65 in the comparison experiment with the real label image. This method effectively overcomes the interference of defect segmentation caused by irregular wavy edge contour, blurred edge, and misidentification of notch, and significantly improves the complete extraction ability of wavy edge feature defects of Si3N4 wafers. It provides key technical support for its promotion and application in high‐end manufacturing fields such as semiconductors.