Research On Clustering And Relevance Feedback In 3D CAD Model Partial Retrieval

Vigorous development and wide application of CAD technologyprovides a wealth of digital resources for new product design,and reasonable reuse these resources can effectively shorten product development cycles, reduce product costs and improve product quality. But compared to the overall structure of the model, local area of model is more complex in geometry and topology structure, whose feature descriptor often corresponding to nonlinear data structure such as tree, graph.Existing model retrieval algorithm can not efficiently complete the task of retrieving a specific model local area. In addition, the existing model retrieval algorithm can barely satisfy the high demand of efficiency and users' individuation. In order to meet the requirements of engineering practice, relevant indicators need to upgradeurgent. Aiming at above issues, this paper take improving search efficiency and accuracy rate as the target, studying relevant three-dimensional C AD model local search algorithm, especially clustering and relevance feedback algorithms.The main work includes the following aspects:(1) using a method based on convolution tree kernel to complete the similar evaluation process for measuring the distance between two models local area, this distance can be used as an important basis for local retrieval,model clustering and relevance feedback.(2) using a method based on figure clustering to complete the model clustering process for non- linear characteristics so as to achieve effective division and integration of model library. At the same time improvingrecall-precision rate, the algorithm can also provide foundation for the extraction of "design pattern" in three-dimensional model field.(3) using a method based on kernel and One-C lass SVM to complete the relevance feedback process so as to achieve users' customization model local search.(4) creatively put forward the "design mode" concept in three-dimensional modeling field, and gave its modeling, extraction and application process.Based on the above research findings, we embedded the local area similar evaluation module, the model clustering module, the relevance feedback module, "design pattern" extraction and label module to the three-dimensional C AD model retrieval platform framework, and gives the corresponding results to verify the main idea of this article.