With the increasing application of Cone Beam Computed Tomography (CBCT) in the dental field, deep learning models demonstrate significant potential for automatically segmenting anatomical structures. We develop a novel model for image segmentation and classification, aimed at providing accurate 3D models for clinical applications. This model comprises two stages: segmentation and classification. In the first stage, we employ a multi-axial network model to segment the CBCT data. Multiple 2D networks are utilized to process the data from different orientations (axial, coronal, and sagittal planes), and the multi-axial segmentation results are merged based on preset hyperparameters to generate 3D segmentation mask data. In the second stage, we employ a 3D network model to classify the 3D segmentation mask data, and we add an attention mechanism module based on 3D Unet, using the 3D segmentation mask data as prior knowledge to mitigate the semantic discrepancies introduced by skip connections. Our proposed network model was clinically applied, enabling the automatic acquisition of precise three-dimensional jawbone data, thereby assisting in the design of customized titanium mesh.