Research On Geometric Similarity Of Machine Parts By Hidden Markov Model

The number of three-dimensional model has increased rapidly in the past 10 years.How to process,analyze and apply the huge number of three-dimensional model has become the focus of research.Effective reuse of 3D model can greatly shorten the product development cycle and reduced the cost of product design,which involves the similarity classification and retrieval technology of 3D model.The traditional classification method is to classify the three-dimensional model manually.The quality of the classification result depends entirely on the ability of the classifier to understand and grasp the three-dimensional model.There is a strong subjective problem.Machine learning is different from traditional classification methods.This method can make machine automatic learning model target features and classification.It has been widely used in the field of image recognition in recent years.As the intuitive information expression of the model,the two-dimensional image conforms to the human visual sensory system and can be used as the input information of Hidden Markov Model in machine learning.This paper combines machine learning with two-dimensional image to study the similarity classification and retrieval of three-dimensional CAD model.Firstly,extracting two-dimensional isometric projection contour map from three-dimensional CAD model and extracting Hu invariant moment,affine invariant moment and wavelet moment features;secondly,the improved and scaled factor Multi-observation sequence of Baum-Welch algorithm is used to train and recognize the model.Finally,the classification and retrieval results of the model are given.The main research work is as follows:(1)Generation of two-dimensional Projection Profile of three-dimensional CAD Model and Feature Extraction of Invariant MomentThe feature of two-dimensional image is the description information of three-dimensional model.The common method of feature extraction is light field descriptor.There is a lot of redundancy in the eigenvalues extracted by this method.In this paper,three-dimensional CAD model is pre-processed by coordinate transformation and used isolateral projection transformation to get the projection image of three-dimensional model.Then,Sobel operator is used to extract the edge contour of the image,so that the two-dimensional equilateral axonometric projection contour map can be obtained,which is used as the geometric structure expression of the three-dimensional model.Using Hu invariant moment,affine invariant moment and wavelet moment theory,the feature extraction of two-dimensional equilateral projection contour map is carried out.The eigenvalues of the three moments,the mixed moments formed between them and the mixed moments formed by the three moments are used as the input observations of Hidden Markov Model to train and recognize the model.In the aspect of feature extraction,the classification and retrieval accuracy of 3D CAD model is improved by utilizing the advantages of various moments.(2)Constructing Hidden Markov ModelHidden Markov Model has strong ability of pattern recognition and description.In this paper,improved Baum-Welch algorithm is applied to Hidden Markov Model.The algorithm consists of two parts: model training and model recognition.Firstly,the eigenvalues of some models are extracted and trained as input observations of Hidden Markov Model.The classification pattern of the optimized model is obtained.Then,the eigenvalues of the remaining models are inputted into the trained optimization model for similarity checking and recognition.Finally,the Hidden Markov Model is used for similarity retrieval and recognition.Model classifier effectively classifies each 3D model and obtains the correct category of each model.Experiments show that the Hidden Markov Model recognition method based on machine learning theory can better solve and realize the learning problem of threedimensional CAD model,and its recognition rate is better than the traditional similarity discrimination and classification algorithm based on Euclidean distance.(3)Evaluation of Model Retrieval PerformanceIn order to investigate the degree of similarity between retrieval results and actual expectations,the feature extraction and classifier algorithm proposed in this paper are analyzed by Recall Precision Curve.The results show that the proposed algorithm is feasible and effective,which provides a new idea for similarity classification and retrieval of CAD model.