简化PCNN在磨粒图像颜色特征提取中的应用

Application of Simplified PCNN in Color Feature Extraction for Wear Particle Images

提出了一种简化脉冲耦合神经网络(pulse coupled neural network,PCNN)的磨粒图像颜色特征提取方法,将简化PCNN分别作用于RGB磨粒图像的3个颜色分量,利用PCNN的脉冲耦合特性对颜色分量进行二值分解。通过定义图像像素捕获率把PCNN二值图像序列转化为一维时间序列,提取捕获率时间序列的均值、标准差、峰值、能量和熵作为磨粒图像的颜色特征,并将提取的特征输入最小二乘支持向量机(least square support vector machine,LS-SVM)对磨粒进行识别。研究结果表明:与单纯的颜色特征相比,PCNN颜色特征能更好地表达磨粒的颜色特征,平均识别率达到98%。

To effectively describe wear particle color characteristics,a color feature extraction method based on simplified pulse coupled neural network （PCNN） was proposed. The method applied simplified PCNN to three color components of every RGB wear particle images respectively and each color component was decomposed into a binary image sequence utilizing PCNN, and the sequences were converted to one dimensional time sequences by defining pixel capture rate. Then mean value, standard deviation, peak, energy and entropy of capture rate sequences were extracted as wear particle image color features. When the extracted features were input into least square support vector machine （LS-SVM） for wear particle recognization,average recognizabilit） reached 98%. Results indicate that the PCNN color features can better represent wear particle color characteristics compared with simplicial color features.