基于支持向量机和方差的管道内表面粗糙度等级识别

针对管道内表面粗糙度等级的非接触式检测问题，提出了一种新的基于支持向量机（SVM）和方差的组合分类方法．SVM是近年发展起来的具备较高分类性能和容噪能力的机器学习方法，但当输入数据量大时，SVM分类的时间耗费太大，系统难以实用化．故本方法首先根据统计方差对待测管道的内表面粗糙度进行分类，再利用SVM进行细分．这样就有效利用了支持向量机识别率高、容噪能力强和统计方差速度快的优点．实验表明本方法具有较好的识别精度、效率和容噪性能．如适当调整参数，本方法还可用于其它物体表面的粗糙度检测，具备良好的推广性．

SURFACE ROUGHNESS DETECTION OF INNER WALL OF PIP BASED ON SVM AND VARIANCE

In this paper, an novel non-contacted approach is presented to measure the inner wall roughness of pipe. The approach is based on gray-level variance and Support Vector Machine (SVM) which considers a candidate to replace neural networks and other traditional classification methods for its good performance. But the time consumption of SVM is vast when the input data is vast. A referance is made to other related papers and our experiments, the surface of a measured specimen can be classified coarsely according to its gray-level variance. Then the surface roughness can be detected by SVM. The approach takes advantages of both SVM and variant. The experiments show that the approach is accurate and efficient. Furthermore, it can be used widely in detecting the surface roughness of other objects by adjusting some parameters rightly.