瓶装食用油异物检测算法

Impurity Detection Algorithm for Bottles Filled with Edible Oil

现有的基于旋转-急停平台上序列图像分析的瓶装液体异物检测方法不适用于食用油生产线。文章提出一种基于单帧图像的瓶装食用油异物检测算法。首先利用均值漂移算法实现食用油的彩色图像分割,提取油体兴趣域;其次,利用积分图像快速估计油体区域的光强分布,并对甁壁高光、暗黑背景造成的干扰进行光强误差校正;最后结合异物区域相较油体区域偏暗的准则实现异物检测。在实际生产线上的实验表明,所提算法能够自适应瓶身以及瓶间不同的光照变化,检测正确率高。

Existing impurity detection algorithm based on analysis of sequencial images captured from rotating-scramimg platform for bottles filled with liquid is not applicable for edible oil production lines. In this paper,an impurity detection algorithm for bottles filled with edible oil using a single image is proposed. First of all,mean-shift algorithm is used to segment the edible oil image,and region of interest is extracted. Secondly,the illumination distribution is fast estimated using integral image technique,and interference errors caused by highlights at the walls of the bottles and dark backgrounds are corrected. Finally,impurities are detected based on the principle that the image areas of foreign substances are darker than the oil regions. Experiments on the factual production lines show that the presented algorithm is adaptive to illuminant variations on the same bottle or between different bottles,and has high detection accuracy.