The Research On Electrode 3D Model Classification And Retrieval Based On SVM And Shape Features

The rapid development of computer hardware, 3D modeling software and internet technology,as well as the advantages of describing objects in real world by 3D model in natural and visual away, have greatly promoted the widely use and popularization of 3D model in many fields, such as mechanical manufacture, video games and virtual reality.In the field of mechanical industry, the traditional 2D CAD has been replaced by 3D model technology. Huge amounts of 3D models and processing technic have being created and saved in the data base. Using the exiting similar 3D model and its relative knowledge to design a new part and its process plan will increase the design efficiency and reduce mistakes. So how to classify the created model, and find out similar 3D model and processing technic quickly and correctly from a 3D model library has become a issue of design. The main content of this thesis is to realize the 3D model retrieval and SVM based classification.In this thesis, two methods of shape feature extraction both based on invariable moment were applied to construct the SVM vector which describes the 3D model shape feature. Classification criterions of electrode 3D model were put forward from shape and NC machining process. A new feature extracting method based on face element work was proposed, which had been proved from the theory and experiment to have the rotation, translation and scale invariance. Experiments of similarity measure and electrode 3D model retrieval using above extracting methods were conducted, the results show that all of them achieve satisfactory retrieval results, and the improved method has better generalization performance. Many experiments were carried out to show how to select best SVM, train classification model and predict models using the k-fold cross validation and"grid-search"methods. The experiments indicate that the classifications of six and two classes of electrode reach the accuracy of 88.6148% and 95.1531% respectively. Finally, a 3D model retrieval & classification system based on above theories was established. It shows that the system has a good classification and retrieval accuracy.