灌装液体药品质量的机器视觉检测与识别

来源期刊：中南大学学报(自然科学版)2009年第4期

论文作者：何 成 王耀南

文章页码：1003 - 1007

关键词：视觉检测；序列图像处理；支持向量机；杂质识别

Key words：visual detection; serial images processing; support vector machine; impurity recognition

摘    要：针对机器视觉检测灌装液体药品中存在的对噪声敏感、难以区分杂质和气泡的问题，提出多帧序列图像检测与识别杂质的方法。在获取灌装液体药品连续序列图像的基础上，采取以最不相邻图像差分来获取瓶内液体图像中变动区域；选取多帧序列图像中运动对象建立目标链，从而提取可能是杂质的运动对象在多帧序列图像中的特征。运用具有较强推广能力的支持向量机对提取特征进行分类识别。实验结果表明：采用该方法可识别灌装液体药品的杂质，选取合适的支持向量机惩罚参数和核参数，其识别准确率可达95%，可以区分杂质和气泡等，可提高检测灌装液体药品质量的准确性。

Abstract: Based on the fact that in detecting impurities in liquid medicine using machine version, there are some problems such as high sensitivity to noise and difficulty in distinguishing impurity and air bubbles, the method using multi-frame serial images was proposed. The serial images of the liquid medicine were obtained, the least-adjacent-image difference algorithm was used to detect the changing regions in the images. Since the moving targets could be traced in the multi-frame serial images, the characteristics of these targets, which were also probably the impurities, could be derived. Support vector machine with high generalization capability was used to categorize and detect these characteristics. The experiment results show that with this multi-frame serial images analysis and appropriate penalty parameter and Kernel-parameter of support vector machine, the detection accuracy can reach 95%. Impurity identification in small sample liquid medicine, the problem of distinguishing impurity and air bubbles can be solved, and high accuracy of detection can be achieved.

基金信息：国家自然科学基金资助项目
国家“863”计划项目