人工智能技术在烟草包装印刷质量检测上的应用研究  

Application Research of Artificial Intelligence Technology in Packaging and Printing(small sheet)Detection

作　　者：钱隽 QIAN Jun(Shanghai Tobacco Packaging and Printing Co.,LTD.,Shanghai 200137,China)

机构地区：[1]上海烟草包装印刷有限公司,上海200137

出　　处：《绿色包装》2020年第11期34-39,共6页Green Packaging

摘　　要：人工智能的发展日新月异,正推动着印刷行业质量检测技术的革新。在总结传统检测算法不足的基础上将深度学习理论应用到该检测领域中,并在实际应用中针对烟草小盒包装印刷产品存在较高误检率的问题,将无监督和二分类有效结合。首先通过无监督模式进行待检产品与好品训练集的严格一致性界定,然后将异常包装印刷产品根据其缺陷子图进行二分类划分为误检品和缺陷品,并以二分类结果作为最终检测结果,最终实现高检出率、低误检率的目标。针对烟草包装印刷小盒产品进行多组实验,以GoogLeNet网络模型为基础,以小张包装印刷品为检测对象,以多种光学成像方式,验证上述无监督模式结合二分类的深度学习策略具备良好检测效果。The rapid development of artificial intelligence is driving the innovation of technology in the printing inspection industry.On the basis of summarizing the shortcomings of traditional detection algorithms,deep learning theory is applied to this detection field,and in practical applications,there is a high false detection rate for small sheet packaging and printing products.Unsupervised mode and two classifiers are effectively combined.Firstly,the strict consistency of the product to be detected and the good product training set is defined by unsupervised mode,and then the abnormal packaging and printing products are divided into two products according to their defect sub-pictures:misdetected products and defective products,and the results of the two classifiers are used as the final detection results.Eventually,achieve the purpose of high detection rate and low false detection rate.Multiple sets of experiments have been carried out on small sheet packaging and printing products,based on the GoogLeNet network model,using small sheets of printed products as the detection object,and using multiple optical imaging methods to verify the above unsupervised mode combination of two classifier of deep learning strategies has a good detection effect.

关 键 词：人工智能技术 深度学习 包装印刷检测 缺陷检测 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TS8[自动化与计算机技术—控制科学与工程]