【作者】 刘帆帆； 贾克斌； 李志坚；

【Author】 LIU Fanfan;JIA Kebin;LI Zhijian;School of Information and Communication Engineering,Beijing University of Technology;

【机构】 北京工业大学信息与通信工程学院；

【摘要】 如何采用新一代信息处理技术来提高生产线自动化检测水平是各个国家重点关注的内容之一。针对目前制药行业广泛存在的人工检测泡罩包装药品缺陷时所出现的效率低、漏检率及成本高等问题,综合运用计算机视觉、深度学习等技术对铝塑泡罩包装药品进行实时检测。对采集到的图像进行预处理,利用仿射变换去除不相关背景,获取缺陷可能存在的感兴趣区域。利用阈值分割、形态学操作,构建不同缺陷类型的特征算子,比较待测和正常药板轮廓、区域特征值的差异,对待测药板缺陷进行分类。利用YOLOv5网络在人工标注的数据集上先对胶囊进行预训练,使用得到的权重对药板继续训练,最终使用训练好的模型对感兴趣区域存在的缺陷进行检测。实验结果表明,算法可以实现对铝塑泡罩包装缺陷准确识别,平均检测精度达到了0.95以上,能够满足泡罩缺陷在线检测的要求。更多还原

【Abstract】 How to use the new generation of information processing technology to improve the automatic detection level of production line is one of the key concerns of various countries. Aiming at the problems of low efficiency, missed detection rate and high cost in manual detection of blister-packaged drug defects in the pharmaceutical industry, computer vision, deep learning and other technologies are combined to perform real-time detection of aluminum-plastic blister-packaged drugs. Firstly, the collected images are preprocessed, and affine transformation is used to remove irrelevant backgrounds to obtain regions of interest where defects may exist.Secondly, threshold segmentation and morphological operations are used to construct feature operators of different defect types, compare the differences between the contours and regional eigenvalues of the to-be-tested and normal drug plates, and classify the defects of the to-be-tested drug plates. Finally, the YOLOv5 network is used to pre-train the capsules on the manually labeled data set, and the obtained weights are used to continue training the medicine plate, the trained model is used to inspect the defects in the region of interest. The experimental results show that the algorithm can accurately identify the defects of aluminum-plastic blister packaging,and the average detection accuracy can reach more than 0.95, which can meet the requirements of online detection of blister defects.更多还原