磨粒图像的纹理分析及识别

Texture Feature Extraction and Recognition of Wear Debris Image

以磨粒显微图像分析为应用背景,引入方向测度对磨粒图像表面纹理特征进行描述.该方法对磨粒图像各方向的灰度变化规律进行统计分析,提取了8个纹理特征.然后以提取的纹理特征为输入矢量,利用径向基函数神经网络对磨粒纹理进行分类识别.应用实例表明,方向测度综合反映了磨粒纹理的方向性和粗糙性,可用于磨粒纹理特征的描述;所建立的基于神经网络的磨粒纹理分类模型学习速度快,识别率较高.

A method and results of wear debris texture description and classification were presented. Direction measure was used to describe the microscopic wear debris texture, and eight texture features were extracted. Used as the input vector of RBF neutral network, the wear debris was divided into four classes: smooth, rough, striation and pitted. The result of application shows that direction measure describes both the orientation and roughness of wear debris surface. And the classification system based on neutral network is fast in convergence, and high in accuracy.