CT Image Automatic Recognition Method Based on Two-Dimensional Segmentation and Three-Dimensional Visualization

To address defect detection challenges in ceramic matrix composite additive manufacturing, this study proposes a CT defect recognition method integrating two-dimensional U-Net segmentation with three-dimensional surface rendering. The framework achieves high-precision segmentation of single-layer CT images via U-Net, followed by defect distribution modeling using the Marching Cubes algorithm. Experimental results demonstrate that the method significantly reduces GPU memory consumption while maintaining spatial localization accuracy. Defect distributions are visualized through color-coded surfaces, meeting engineering inspection requirements for large-scale additive manufacturing components. Compared to traditional 3D networks, this approach balances segmentation accuracy and computational efficiency, offering an efficient quality inspection solution for ceramic matrix composite additive manufacturing production.