In order to reveal the tensile damage evolution and failure mechanism of plain weave Cf/SiC composites, the X-ray CT in-situ tensile test was carried out to obtain the three-dimensional reconstructed image of the material. The deep learning based image segmentation method was used to accurately identify the tensile crack and realize its three-dimensional visualization. The damage evolution and failure mechanism of plain weave Cf/SiC composites were analyzed, and the damage was quantitatively characterized based on the three-dimensional visualization results. The results show that the tensile mechanical behavior of plain weave Cf/SiC composites is nonlinear, and damages such as matrix cracking, interface debonding, fiber fracture, and fiber pull-out occur during the tensile process. The initial defects are easy to cause material damage, and the higher the porosity, the more cracks there are. The matrix crack outside the fiber tow can extend to the inside of the fiber tow and the crack deflection occurs. The deep learning based intelligence image segmentation method provides an effective analysis method to quantitatively evaluate the damage evolution and failure mechanism of ceramic matrix composites.