论文部分内容阅读
为解决条烟装箱过程中出现的烟箱缺条等质量问题,基于烟箱内部前位、后位图片,建立了一种基于相似性分析和阈值自校正的烟箱缺条智能检测方法.首先,将前位和后位图片进行灰度变换和图像分割,基于处理后图片的横向和纵向分块,采用皮尔逊相似性分析计算图片的16个特征相似度指标,利用核密度估计确定所有建模图片16个指标的阈值;其次,采用横向和纵向分块的相似性分析方法,计算前位和后位图片的16个特征相似度指标,通过与阈值的比较识别前位和后位图片是否发生缺条;最后,将新的未缺条图片特征相似度指标添加到建模图片的特征相似度矩阵中,利用核密度估计进行阈值自校正.以杭州卷烟厂条烟装箱过程的实际运行数据进行验证,结果表明:该方法实现了烟箱内部图片特征信息的有效表征和量化,能够准确判断烟箱缺条缺陷,对测试数据中所有15种缺条缺陷的检测准确率为100%,标注后的图片为缺条问题追溯提供了依据.该方法可为提高烟箱缺条检测方法的适应性和可靠性提供技术支持.“,”In order to detect cigarette carton missing in cigarette case, an intelligent detection method based on similarity analysis and threshold self-correction was proposed on the basis of the front position and rear position images inside cigarette case. Firstly, the front position and rear position images were processed by gradation transformation and image segmentation; based on the horizontal and vertical segmentation of the processed images, 16 feature similarity indicators of the processed images were calculated by Pearson similarity analysis; and the thresholds of the 16 indicators for all modelling images were determined by kernel density estimation. Secondly, the 16 feature similarity indicators of the front position and rear position images were calculated by horizontal and vertical block similarity analysis method, and carton missing in the current front and rear images was judged via comparing with the thresholds. Finally, the feature similarity indicators of the new without carton missing images were added to the feature similarity matrix of the modelling images, and threshold self-correction was carried out with kernel density estimation. Off-line validation was carried out on the basis of actual running data of cigarette case filling process in Hangzhou Cigarette Factory, the results showed that the proposed method realized the effective characterization and quantification of feature information of the images inside cigarette cases and could accurately detect the carton missing in cigarette cases. The detection accuracy reached 100% for all 15 types of carton missing in the test data, and the labeled images provided the basis for carton missing tracing. The proposed method provides a technical support for improving the adaptability and reliability of detection method for carton missing in cigarette case.