基于比例特征的区域分割算法在药板缺陷检测中的研究

Region Segmentation Algorithm Based on Proportional Features for Defect Detection of Aluminum Plastic Blister Medicine Plates

中文关键词: 铝塑泡罩药板  比例特征  缺陷检测  归一化互相关

英文关键词: aluminum plastic blister medicine plates  proportional characteristics  defect detection  normalized cross correlation (NCC)

中文摘要:

      目的 为解决铝塑泡罩药板图像ROI区域定位慢、精度差等问题，本文提出一种基于比例特征的泡罩区域分割算法，该算法可以快速定位并分割泡罩ROI区域，结合图像相关性特征算法对铝塑泡罩药板进行缺陷检测。方法 首先通过工业相机采集药品包装生产线上的药板原始图像，接着使用Blob分析从原始图片中分离出铝塑泡罩主体部分，然后通过仿射变换将图像放置在中心区域，并使用比例特征分割算法对泡罩区域进行分割，最后通过金字塔加速的NCC算法完成缺陷检测。结果 实验结果表明，基于比例特征分割后的图像平均NCC匹配时间为9 ms，在缺陷样本占比20%的实验中误检率为0.167%，漏检率为0.556%。结论 通过比例特征分割出精准的泡罩ROI区域结合改进的NCC算法，在拥有较高准确率的同时大幅减少了缺陷检测时图像匹配的时间，能较好地完成铝塑泡罩药板的缺陷检测任务。

英文摘要:

      The work aims to propose a blister area segmentation algorithm based on proportional features to quickly locate and segment the blister ROI, and detect defects in aluminum plastic blister medicine plates in combination with the image correlation feature algorithm, so as to solve the problems of slow localization and poor accuracy of ROI in images of aluminum plastic blister medicine plates. Firstly, original images of medicine plates in the packaging production line were collected through an industrial camera. Then, Blob analysis was used to separate the main part of the aluminum plastic blister from the original image. Then, the image was placed in the center area through affine changes and the blister area was segmented according to the proportional feature segmentation algorithm. Finally, defect detection was completed according to the pyramid accelerated NCC algorithm. The experimental results showed that the average NCC matching time of the image based on proportional feature segmentation was 9 ms. In the experiment with 20% defect samples, the false detection rate was 0.167% and the missed detection rate was 0.556%. By the segmenting precise blister ROI through proportional features and combining them with an improved NCC algorithm, the image matching time during defect detection is significantly reduced, which can effectively complete the defect detection task of aluminum plastic blister medicine plates.