The invention provides a multi-layer information sharing and correcting method for an XCT slice, which is a post-processing algorithm based on a full convolutional neural network and can realize high-precision identification of a microstructure of the XCT slice of a woven ceramic matrix composite material. Wherein the input of the network is a single arbitrary ceramic matrix composite XCT slice with the same resolution, and the output of the network is a semantic segmentation result of a mesoscopic structure. After the segmentation is completed, the common point of the information between theadjacent slices is used for correcting the wrong pixel category of the current slice, so that the continuity of the label is enhanced, and the accuracy of the segmentation is improved.