Dental model point cloud segmentation method based on cross-graph attention mechanism and cost function learning

Abstract

The invention discloses a dental model point cloud segmentation method based on a cross-graph attention mechanism and cost function learning. The method comprises the steps: firstly constructing a dental model point cloud segmentation model, building an interaction graph network of heterogeneous geometric data in the model, and exploring local information in the same adjacent graph and between different adjacent graphs through the cross-graph attention mechanism. The dependency among the heterogeneous geometric data is learned, the identification capability of context sensing features is improved, and the problem that current heterogeneous geometric data analysis only analyzes each type of data or simple linear combination heterogeneous data neglects a semantic gap among the heterogeneous data is solved; according to the method, on the basis of NAS (NeuralArchitecture Search), a cost function is designed through automatic machine learning and an evolutionary algorithm, the cost function is formulated as an original mathematical operator of a tree structure, the cost function with the highest consistency with a metric function is solved, and the problem that the cost function is inconsistent with the metric function is solved. Compared with other advanced methods, the method has considerable competitiveness.

Description

technical field [0001] The present invention relates to a point cloud segmentation technology, in particular to a dental model point cloud segmentation method based on a cross-graph attention mechanism and cost function learning. Background technique [0002] Digital dental technology has developed rapidly in recent years. Dental model point cloud segmentation obtains each tooth area from a 3D model built by intraoral or desktop scanning, which plays an important role in digital dentistry and can be used in applications such as orthodontic diagnosis, oral surgery, and treatment planning. Recently, significant progress has been made in deep learning-based 3D segmentation algorithms. However, heterogeneous geometric data are analyzed individually or linearly combined, for example, in a two-stream graph convolutional network (TSGCNet), the 3D coordinates and normal vectors of a triangular mesh are analyzed via C-Stream and N-Stream, respectively, ignoring heterogeneous Semant...

AI Technical Summary

Problems solved by technology

However, heterogeneous geometric data are analyzed individually or linearly combined, for example, in a two-stream graph convolutional network (TSGCNet), the 3D coordinates and normal vectors of a triangular mesh are analyzed via C-Stream and N-Stream, respectively, ignoring heterogeneous The Semantic Gap Between Qualitative Data

In addition, there is an inconsistency between the cost function and the metric function in the segmentation task, specifically, the cross-entropy is used to minimize the point-wise classification error, while the mean intersection-over-union (mIoU) is used to measure the segmentation results , which is not directly optimized during the training phase due to the non-differentiable nature of mIoU

Images