The image collection of bearing surface defects is conducted by using CCD instead of eyes. The image processing method that combines convolution filtering with open, close operation effectively removes interferences coming from the edge points around the defects. The feature quantities are increased based on the traditional features, such as degree of compression, line length, distance extremes ratio, NMI feature and invariant moments, the basis is enhanced for classification of defects. The input matrix of BP neural network and normalized method is improved, and the memory capacity and recognition speed of neural networks are advanced; The reliability of the system is versified by detecting the recognition result of defect classification.%用CCD代替人眼对轴承表面缺陷进行图像采集,采用卷积滤波与开、闭运算相结合的图像处理方法,有效去除了缺陷周围边缘点的干扰.在提取传统特征基础上增加了压缩度、线度、距离极值比、NMI特征和不变矩等特征量,增强了缺陷分类的依据;对BP神经网络的输入矩阵和归一化方法的改进,提高了神经网络的记忆能力及识别速度;通过试验对缺陷分类系统识别结果进行检测,确定了该系统的可靠性.
展开▼