中文摘要：

针对生产和装配过程中轴承表面缺陷检测传统方法的不足,提出一种新的轴承表面缺陷类型识别算法。首先改进Canny算子以提高轮廓识别度,将Sift算法应用于缺陷区域提取,对轴承表面缺陷图像和无缺陷图像进行Sift图像匹配以定位缺陷区域,运用像素点的异或运算以精确提取缺陷区域。选择部分Hu矩值和几何特征值准确描述缺陷区域,将其作为BP神经网络算法的输入,从而最终识别出缺陷类型。实验表明,该方法提高了识别率,且具有非接触、速度快、精度高和抗干扰能力强等优点,较好地实现了轴承表面缺陷类型的检测。

英文摘要：

Aiming at some shortcomings in the traditional recognition methods for the bearing surface defect generated during the process of production and assembly,this paper presented a new bearing surface defect recognition algorithm. Firstly,it put forward an improved Canny operator to enhance the recognition rate,and also applied Sift image matching algorithm on the bearing surface defect extraction to locate the defect area with or without defect by matching the images. It used the pixel XOR operation to extract the defect area precisely and selected part of Hu moment and geometric features values to describe the defect area accurately and used as the input data for the BP neural network algorithm. Finally,it identified defect type. Experiments show that this method improves the recognition rate,and with the merits of non-contact,fast speed,high accuracy and strong anti-jamming capability,so it can realize the recognition of bearing surface defect type accurately.