基于机器视觉的硅锭端面隐裂检测方法

Silicon Ingot End Face Crack Detection Method Based on Machine Vision

摘要:

硅锭在生产过程中难以避免有端面隐裂的发生,严重影响生产线的安全性能和生产效率,为此提出一种改进背景差分算法的硅锭端面隐裂自动检测方法和系统。首先,分析端面隐裂的特点,采用中值滤波来对图像进行预处理;其次,采用图片样本库训练得到背景图像并对常用差分算法进行改进，在对图像进行差分的同时增强隐裂缺陷；再根据隐裂缺陷的最大比例不超过10%进一步去除差分图片中的大部分背景点；最后再提取隐裂缺陷区域,对区域进行连通域分析并根据缺陷区域面积来判断端面是否隐裂。仿真结果和现场实践表明,该方法对硅碇隐裂的识别准确率达96.4%,召回率达98.5%，识别时间低于200ms,很好地满足现场运行的需求。

Abstract:

It is difficult to avoid the occurrence of end face crack of silicon ingot in the production process, which seriously affects the safety performance and production efficiency of the production line. Therefore, an improved background difference algorithm for automatic detection of end face crack of silicon ingot is proposed. Firstly, the characteristics of the end face crack are analyzed, and the median filter is used to preprocess the image; Secondly, the background image is trained by the image sample library and the commonly used difference algorithm is improved to enhance the crack defect while performing the difference on the image; Then, most of the background points in the difference picture are further removed according to the maximum proportion of crack defects not exceeding 10%; Finally, the cracked defect region is extracted, and the connected region is analyzed to determine whether the end face is cracked according to the area of the defect region. The simulation results and field practice show that the recognition accuracy of this method is 96.4%, the recall rate is 98.5%, and the recognition time is less than 200ms, which meets the needs of field operation.