基于人工免疫算法的印刷品缺陷检测技术研究

发布时间：2018-03-29 21:19

本文选题：印刷品缺陷检测　切入点：人工免疫　出处：《西安理工大学》2017年硕士论文

【摘要】：在印刷行业中,印刷品缺陷检测对印刷品质量的评定与控制具有重要意义,在实际生产中,要剔除带有缺陷的印刷品。现阶段的印刷品缺陷检测方法大都存在着或多或少的不足,算法有待更新。图像处理技术的发展和新的智能算法的出现促进了检测技术的进步;多种智能算法的交叉融合应用,有望开发出一种新的更高效的印刷品缺陷检测方法。本文基于人工免疫算法对印刷品缺陷检测技术进行了研究。人工免疫算法具有多层检测机制,需要少量或不需要先验知识,仅需要少量缺陷样本的优点。因此,基于人工免疫算法的印刷品缺陷检测技术具有广阔的发展空间。在研究了人工免疫算法的阴性选择原理的基础上,结合图像处理技术,利用能够反映图像像素灰度分布规律的灰度共生矩阵,求得印刷品图像在0°、45°、90°和135°四个方向上的能量、熵、对比度、相关性、同质性五个纹理特征,并将各特征值的均值作为图像特征的最终值,以各特征值的均值组成特征向量作为阴性选择算法的数据表示。本文研究了基本阴性选择算法并对其进行了改进,提出实数值编码的改进阴性选择算法,样本数据空间采用多维实数值向量表示;利用欧氏距离计算两样本间的亲和度,并判断它们是否发生匹配。在缺陷检测方面,针对基本阴性选择算法只能识别正常和缺陷,而不能检测缺陷种类的特点,提出引入“疫苗”的改进办法,对缺陷类型进行识别。该方法提取已知缺陷印刷品图像特征向量作为“疫苗”,对检测器进行聚类操作,构造多种检测器集合,采用多检测器集融合诊断对缺陷种类进行判断。为了提高算法的检测精度,降低误判率,本文提出将待检样本与自己集合进行二次匹配的方法。最后,本文基于MATLAB编程实现了从图像处理到缺陷检测的全部过程,设计了印刷品缺陷检测系统。实验结果表明,本文提出的改进阴性选择算法能够有效地检测出印刷品缺陷;设计的印刷品缺陷检测系统能够快速识别出缺陷种类,且检测结果较为准确,有一定的应用价值。
[Abstract]:At this stage, most of the defect detection methods of printed matter have more or less shortcomings, and the algorithm needs to be updated.The development of image processing technology and the emergence of new intelligent algorithms promote the progress of detection technology, and the cross-fusion application of many intelligent algorithms is expected to develop a new and more efficient method of print defect detection.In this paper, based on artificial immune algorithm, print defect detection technology is studied.The artificial immune algorithm (AIA) has the advantages of multi-layer detection, which requires little or no prior knowledge, and only a small number of defect samples.Therefore, the printing defect detection technology based on artificial immune algorithm has a broad development space.On the basis of studying the principle of negative selection of artificial immune algorithm and combining with image processing technology, the energy and entropy of printed image in four directions of 0 掳~ 45 掳~ 90 掳and 135 掳are obtained by using the gray level co-occurrence matrix which can reflect the law of image pixel gray distribution.Contrast, correlation and homogeneity are five texture features. The mean value of each eigenvalue is taken as the final value of the image feature, and the average value of each eigenvalue is used as the data representation of the negative selection algorithm.In this paper, the basic negative selection algorithm is studied and improved. An improved negative selection algorithm based on real value coding is proposed. The sample data space is represented by multidimensional real value vector, and the affinity between two samples is calculated by Euclidean distance.And determine whether they match or not.In the aspect of defect detection, in view of the fact that the basic negative selection algorithm can only recognize the normal and the defect, but can not detect the type of defect, an improved method of introducing "vaccine" is put forward to identify the type of defect.In this method, the feature vectors of printed image of known defects are extracted as "vaccines", the detectors are clustered, the sets of multiple detectors are constructed, and the types of defects are judged by the fusion diagnosis of multi-detector sets.In order to improve the detection accuracy of the algorithm and reduce the error rate, this paper proposes a method of quadratic matching between the samples to be checked and their own sets.Finally, the whole process from image processing to defect detection is realized based on MATLAB programming, and the print defect detection system is designed.The experimental results show that the improved negative selection algorithm proposed in this paper can effectively detect print defects, and the designed print defect detection system can quickly identify the types of defects, and the detection results are more accurate and have certain application value.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TP391.41;TS807

【参考文献】