摘要
提出了一种基于机器视觉的分布式浮法玻璃缺陷在线检测识别方法。针对采集到玻璃图像为非均匀图像的特点,首先对图像进行平滑滤波处理,剔除信号波动带来的灰度尖峰和毛刺,然后设计了基于自跟踪的向下阈值曲面分割方法实现缺陷和背景的分割,实现了基于扫描线的图像搜索算法保证缺陷寻找的速度,并利用神经网络对缺陷进行分类。实际应用表明,该系统能够准确可靠的实现在线检测和识别浮法玻璃缺陷,满足了浮法玻璃企业对玻璃缺陷实时检测的需要。
An online distributed system for float glass defect inspection is presented. Firstly, in allusion to the non -uniformity float glass image, the glass image is treated by smooth filtering so as to eliminate the pinnacles caused by signal fluctuation. Secondly, a downward segmentation method based on self- track threshold surface is presented to separate the defects and background. Thirdly, a scanning beam image searching algorithm is realized to decrease the defect searching time. Lastly, a three- neural network classification is designed for float glass defect recognition. The practical application shows that the online inspection system based on this method can satisfy the requirements of defect inspection and recognition for float glass fabrication.
出处
《玻璃与搪瓷》
CAS
2010年第1期1-6,共6页
Glass & Enamel
基金
武汉市科技攻关计划资助项目(06-7)
关键词
浮法玻璃
缺陷检测
阈值曲面
神经网络
float glass
defect inspection
threshold surface
neural network