数码产品外壳表面的适应性智能检测技术

Adaptive Intelligent Detection Technology for Digital Products' Shell Surface

针对不同的数码产品表面光学特性不一、缺陷检测可靠性低的问题,提出了一种可适应不同表面类型的视觉检测方法.首先在不同的光源下采集图像,并根据灰度统计分析后的识别结果对材料进行分类;接着使用基于全局和动态阈值的分割算法以及改进的曲线检测器对不同的表面进行检测,该曲线检测器利用高斯滤波和偏导特征先找出曲线的关键点,再通过"松弛"算法将其连接成线.实验结果表明,文中算法具有较好的鲁棒性,对外界有较强的抗干扰能力;综合性能分析表明,文中算法的检测误报率低于5%,准确率高于93%,检测速度能满足实际生产要求.

As different types of digital products have different superficial optical characteristics, a visual detection method adaptive to various surface types is proposed to improve the reliability of defect detection. Firstly, after the image collection under different light sources, materials are classified according to the recognition results of gray statistic analysis. Secondly, a hybrid threshold segmentation algorithm, which is on the basis of global and dynamic threshold segmentation techniques, as well as an improved curve detector, which uses Gaussian filter and partial derivative feature to find out the curve＇s key points and then connects the key points into a line through the ＂relaxation＂ algorithm, is used to detect different given surfaces. Experimental results show that the proposed algorithm is highly robust and resistive to external disturbances. Moreover, comprehensive performance analysis indicates that the proposed algorithm produces a false alarm rate lower than 5% and an accuracy rate higher than 93%. Besides, the high detection speed makes the algorithm possible to be applied to actual production.