Abstract: For the detection of geometric parameters of shell-type parts without fixture positioning in machine vision, it is necessary to identify the geometric features of parts in order to plan the detection path. Therefor this paper proposes an intelligent recognition method for geometric features based on supervised machine learning. Firstly, the feature matrix is constructed according to the relation between features to be identified of the shell parts, and then the supervised machine learning algorithm is used to identify these features. An error correction method based on feature uniqueness is proposed to correct the identification errors generated in the classification process. For the research case involved in this paper, there are 4 holes to be identified in the part, and the accuracy of intelligent recognition is up to 100% after 5 supervised trainings.

Key words: supervised machine learning, machine vision, geometric features, decision tree, support vector machine

摘要： 针对在采用机器视觉的无夹具定位的壳体类零件几何参数检测过程中，需要先智能识别零件几何特征以规划检测路径的问题，提出一种基于监督式机器学习的几何特征智能识别方法。利用壳体零件待识别特征的中心位置关系构成特征矩阵，利用监督式机器学习算法进行识别，提出一种基于特征唯一性的纠错方法对分类过程中产生的识别错误进行纠正。对于所涉研究实例，零件共有4个待识别孔，在5次监督式训练后智能识别准确度达100%。

关键词: 监督式机器学习, 机器视觉, 零件几何特征, 决策树, 支持向量机