高速医药自动化生产线大输液视觉检测与识别技术

Visual detection and recognition for medical infusion automatic production lines

中文关键词  药大输液  视觉检测  改进模糊细胞神经网络  异物分类识别

英文关键词  medical infusion  visual detection  improved fuzzy cellular neural network  foreign substances classify and recognition

中文摘要

      药品灌装质量检测是制药过程的一个重要环节, 是药品质量的可靠保证. 针对医药大输液可见异物视觉检测的需求, 研制出基于多视觉的大输液自动化检测识别系统. 首先研究了医药图像的高速高可靠性预处理方法, 有效消除由机械振动和跟踪引起的干扰. 研究了以药液微小异物为目标的改进模糊细胞神经网络图像分割方法, 揭示了液体中异物目标、微粒、气泡等产生机理, 综合分析目标的形态特征、边缘轮廓、运行特征等, 得到各种异物的类型特征以及在序列图像中的动态变化信息. 最后, 使用序列图像的目标特征, 基于支持向量机的AdaBoosting分类算法进行异物识别, 结果证明本文提出的方法检测识别率高, 对工程设备的研制具有重要意义.

英文摘要

      The detection of filling quality for pharmaceutical is an important element in pharmaceutical process, which guarantees the medication security. According to the detection demand of visible foreign substances for glass bottle medical infusion, a high-speed automated foreign substance detection system based on the machine vision consisting of multi-types image acquisition devices is developed. Firstly, a high-speed high reliability preprocessing method for medical infusion images is proposed to eliminate the disturbance caused by mechanical vibration and tracking. Then, an improved fuzzy cellular neural network (IFCNN) is developed for image segmentation to effectively solve the problem of edge detection, which reveals the mechanism of foreign substances, particles as well as liquid bubbles, comprehensively analyzes the morphological boundary of the target operating characteristics, and obtains various characteristics and the dynamic change information in image sequences. Finally, the classification approaches of support-vector-machine (SVM) and AdaBoosting, based on image sequences target characteristics, is employed to recognize multiple tiny insoluble foreign substances. Experimental results indicate that the developed approach is efficient and increases the recognition rate, which is of great significance to the development of engineering equipment.