摘要
曲波变换具有多尺度分析能力,与小波变换相比可更好地表达图像的曲线特征。为有效描述铁谱磨粒的形貌特征,提出一种曲波域图像特征提取方法。利用曲波变换将磨粒图像进行分解,得到不同尺度的曲波系数;根据曲波系数统计分布特点,采用广义高斯分布模型对细尺度和精细尺度曲波系数分布进行建模;提取粗尺度曲波系数的均值、标准差、能量和熵等统计特征,以及细尺度和精细尺度曲波系数的广义高斯分布模型参数描述磨粒特征。将提取的特征用于发动机典型磨粒识别,识别成功率达到了88.9%,表明该方法所提特征能很好地表达铁谱磨粒的形貌特征。
Curvelet transform has the ability of multi-scale analysis and can better express cure features of images com- pared with wavelet transform. To effectively describe morphology characteristics of ferrography wear particles, an image fea-ture extraction method in curvelet domain was proposed. Wear particle images were decomposed by curvelet transform, and curvelet coefficients were obtained. According to statistical distribution characteristics of curvelet coefficients, the detail-scale and fine-scale curvelet coefficients were modeled respectively with generalized Gaussian distribution model. Then, sta-tistical features of mean, standard deviation, energy and entropy of coarse-scale curvelet coefficients were extracted to de- pict the characteristics of wear particles as well as generalized Gaussian distribution model parameters of detail-scale and fine-scale curvelet coefficients. When the extracted features were applied for engine typical wear particle recognition, the recognition rate reached 88. 9%. Experimental results prove that features extracted by the proposed method can better ex- press morphology characteristics of ferrography wear particles.
出处
《润滑与密封》
CAS
CSCD
北大核心
2012年第8期61-65,共5页
Lubrication Engineering
基金
国家自然科学基金项目(50705097)
清华大学摩擦学国家重点实验室开放基金资助项目(SKLTKF09B06)
关键词
铁谱分析
磨粒图像
曲波变换
广义高斯分布
特征提取
ferrography analysis
wear particle image
curvelet transform
generalized Gaussian distribution
feature ex-traction