王胜,吕林涛,杨宏才,陆地.PSO-Gabor-CNN算法在印刷品套印缺陷的检测[J].包装工程,2020,41(5):214-222.
WANG Sheng,LYU Lin-tao,YANG Hong-cai,LU Di.Detection of Overprint Defects by PSO-Gabor-CNN Algorithms[J].Packaging Engineering,2020,41(5):214-222. |
| PSO-Gabor-CNN算法在印刷品套印缺陷的检测 |
| Detection of Overprint Defects by PSO-Gabor-CNN Algorithms |
| 投稿时间：2019-06-09 修订日期：2020-03-10 |
| DOI：10.19554/j.cnki.1001-3563.2020.05.031 |
| 中文关键词: 套印缺陷 Sobel算子 二维Gabor滤波器 粒子群算法 卷积神经网络 |
| 英文关键词: overprint defects Sobel operator 2-dimensional Gabor filter particle swarm optimization convolutional neural network |
| 基金项目:国家自然科学基金（61273271）；2016年度陕西省工业科技攻关项目（2016GY-141）；2017年度西安市科技产学研项目（2017087CG/RC050（XJXY001））；西京学院校级科研基金（XJ160232） |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 二维Gabor滤波器含有多个参数，在印刷品套印缺陷检测中，二维Gabor滤波器使用不同参数增强图像特征的效果差别较大，为了获得二维Gabor在某印刷品套印缺陷检测下的优化参数。方法 在印刷品套印缺陷检测中，提出一种PSO-Gabor-CNN算法，采用Sobel算子对印刷品图像进行边缘检测，以粒子群算法（PSO）对二维Gabor滤波器的中心最大频率kmax、带宽σ、模板窗口window进行参数寻优，处理后的图像与模板图像采用加权欧式距离进行评价。然后用优化后的Gabor滤波器对图像进行滤波，最后采用卷积神经网络（CNN）对印刷品套印缺陷进行检测和分类。结果 通过粒子群算法，确定了二维Gabor中心最大频率kmax为6.0476、带宽σ为0.1444、模板窗口window为27×27取得最佳效果，此时加权欧式距离为1.1927×10-33。卷积神经网络经过70次训练的均方误差为0.0035，测试样本正确率为96.93%。该方法与无数据预处理的BP神经网络（BPNN）、Sobel预处理的BP神经网络（Sobel-BPNN）、无数据预处理的卷积神经网络（CNN）、Sobel预处理的卷积神经网络（Sobel-CNN）对比，表现出了较好的识别效果。结论 该方法可以获取二维Gabor滤波器的较优参数，从而获得较好的滤波效果，将其应用于套印缺陷检测，具有一定的应用价值。 |
| 英文摘要: |
| Two-dimensional Gabor filter contains many parameters, and the effect of two-dimensional Gabor filter using different parameters to enhance image features is quite different in printing overprint defects detection. The paper aims to obtain the optimal parameters of two-dimensional Gabor filter in printing overprint defects detection. In the process of overprint defects detection, a PSO-Gabor-CNN algorithm was proposed. Sobel operator was used to detect the edge of printed images. Particle swarm optimization (PSO) was used to optimize the maximum central frequency 'kmax', bandwidth 'σ' and template 'window' of two-dimensional Gabor filter. The weighted Euclidean distance between the processed images and the template images was evaluated, and then the optimized Gabor filter was used to filter the images. Finally, the Convolution Neural Network (CNN) was used to detect and classify the printing overprint defects. The maximum center frequency of two-dimensional Gabor was 6.0476, the bandwidth was 0.1444 and the window of template was 27×27 by particle swarm optimization. At this time, the weighted Euclidean distance was 1.1927×10-33. The mean square error of convolution neural network after 70 times of training was 0.0035, and the accuracy of test samples was 96.93%. Compared with BP neural network (BPNN) without data preprocessing, Sobel BP neural network (Sobel-BPNN) with Sobel preprocessing, convolutional neural network (CNN) without data preprocessing and Sobel convolutional neural network (Sobel-CNN) with data preprocessing, this method showed better recognition effect. This method can obtain the optimal parameters of two-dimensional Gabor filter and obtain good filtering effect. It has certain application value in overprint defects detection. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
渝公网安备 50010702501716号