基于并查集和约束集合的雪糕棒表面污染检测*

雪糕棒表面的污染缺陷会严重影响其卫生安全,是雪糕棒生产企业质量检测的重要内容,而由于人工检测存在低效率、低精度、卫生隐患等问题,促使雪糕棒表面缺陷的检测已逐步向自动化方向发展。为了解决由于污染缺陷尺寸不定、光照分布不均等造成当前雪糕棒表面污染缺陷检测算法性能不佳的问题,本文在对其进行详细分析的基础上,提出了一种基于并查集和约束集合相结合的检测方案。首先进行图像预处理,根据先验知识对目标雪糕棒进行粗定位,进而通过OTSU算法分割出感兴趣区域；然后,基于并查集算法结合定义的最小区域距离值和灰度差分值将目标雪糕棒表面满足预设阈值条件的像素点合并起来,即将上一步骤预处理所得图像分割成了若干子区域；最后,根据定义的约束集合对各子区域进行筛查以去除其中的噪声区域,从而实现污染缺陷的检测。在自建的图像数据库SUT-I1上进行了算法效果测试,结果表明,本文方法对污染缺陷检测的等误率仅为4.78%,与其它检测方法相比其等误率至少降低了9.44%,体现出本文方法的优越性,具有一定的实际应用价值。

Detection of contamination defects on the surface of the ice cream bar based on the Union-Find Sets and constraint sets

Contamination defects on the surface of the ice cream bar will seriously affect the health and safety, which is an important part of the ice cream bar production enterprise quality detection. Due to the artificial detection such as low efficiency, low accuracy and potential health problems, the ice cream bar surface defect detection has gradually to develop in the direction of automation. In order to solve the problem of the defect size variable and illumination distribution nonuniform which causing the poor performance of the current ice cream bar surface contamination defect detection algorithm, a detection scheme combined with Union-Find Sets and constraint sets is proposed on the base of the detailed analysis. Firstly, the ice cream bar is positioned rudely which belongs to the image pre-processing, then ROI is segmented out via OTSU. Secondly, the pixel points meeting the preset threshold condition on the target ice cream bar are combined based on the Union-Find Sets algorithm combined with the minimum area distance value and gray difference score. Finally, the noise regions are get rid of through screening the subdomains according to the defined constraint sets. The algorithm is tested in the self-built image database. The results show that the contamination defects detecting EER proposed in this paper is only 4.78 percent, which decreases 9.44 percent at last comparing to the other algorithms EER. It indicates the superiority proposed in this paper, which is of actual use value.