文献类型：期刊文献

中文题名：工件装配基准端的形状识别研究

英文题名：Shape recognition research on assembly datum end of workpiece

作者：沈霞[1];郭慧[1];王勇[1]

机构：[1]华东理工大学机械与动力工程学院,上海200237

年份：2015

期号：2

起止页码：97

中文期刊名：现代制造工程

外文期刊名：Modern Manufacturing Engineering

收录：CSTPCD;;北大核心:【北大核心2014】；CSCD:【CSCD_E2015_2016】；

语种：中文

中文关键词：装配基准端;改进的Hu不变矩;BP神经网络;LM算法;识别

外文关键词：assembly datum end ; improved Hu invariant moments; BP neural network; LM algorithm ; recognition

摘要：为了有效识别工件的装配基准端,根据工件两端不同的形状特征,提出了一种基于改进的Hu不变矩和LM-BP神经网络的工件装配基准端识别方法。该方法针对Hu不变矩在离散图像缩放运算上存在较大误差的问题,采用改进的Hu不变矩提取工件两端的形状特征值;用提取的特征值训练LM(Levenberg-Marquardt)算法优化的BP神经网络(即LM-BP神经网络),实现工件两端的形状识别,判断装配基准端。实验结果表明,改进的Hu不变矩能保证特征值在图像缩放情况下的不变性,改进的Hu不变矩与LM-BP神经网络结合的识别算法对工件两端形状具有很好的识别能力。
To recognize the assembly datum end of workpiece effectively. A method of improved Hu invariant moments and BP network is proposed to identify the assembly datum end of workpiece according to the different shape feature of two ends of the workpiece. Aiming at the problem of a huge error in the discrete image scaling operation with using Hu invariant moments, the im- proved Hu invariant moments is used to extract the feature values of the shape of two ends of workpiece, the extracted feature val- ues is used to train the BP network optimized by Levenberg-Marquardt （LM） , to achieve the shape recognition of two ends of workpiece,and the assembly datum end is judged. The experimental results show that the improved Hu invariant moments can guarantee the invariance of feature values in the case of image scaling, and have a good ability of recognition capability to the both ends shape of workpiece combined with the recognition algorithm of the BP neural network.