摘要
提出了基于小波变换提取零件图像特征和用神经网络实现识别的方法。首先,对零件图像进行小波多尺度边缘检测,提取零件图像的边缘轮廓,然后将被检测的边缘轮廓图像分成若干个子区域并分别统计各子区域的边缘像素量,各子区域中的相对边缘像素系数作为零件的特征,将这些特征作为神经网络的输入样本,由神经网络实现识别。3种子区域的不同数量样本的实验结果证明了提出的方法是有效的。
A method to extract part image features and to recognize parts based on wavelet-neural networks was presented. Firstly, the edges from part images were detected,using wavelet multiscale edge detection. Then, edge images were divided into several sub areas and their edge pixels were coun.ted respectively, the relative edge pixel coefficient in each sub-area,was cdnsidered as its feature and finally the features were used as the inputs of a neural network to realize pattern recognition. Ex perimental results using different quantity samples in three kinds of sub-areas volidate that the proposed method can efficiently recognite parts.
出处
《中国机械工程》
EI
CAS
CSCD
北大核心
2005年第22期2031-2033,共3页
China Mechanical Engineering
基金
江苏省教育厅自然科学项目(02KJB470006)
关键词
小波变换
特征提取
模式识别
神经网络
wavelet transform
feature extraction
pattern recognition
neural network
作者简介
夏庆观,男,1947年生。南京工程学院自动化系副教授。研究方向为检测技术。出版专著2部,发表论文13篇。
盛党红,女,1965年生。南京工程学院自动化系副教授、博士。
路红,女,1975年生。南京工程学院自动化系硕士研究生。
陈桂,女,1973年生。南京工程学院自动化系硕士研究生。