基于梯度向量特征提取的安瓿瓶外观检测方法  

Ampoule bottle appearance detection method based on gradient vector features extraction

作　　者：余小游[1] 胡小梅[1] 刘巨昌 陈超艳[1] 王晓春[3] 詹茵茵[1] 

机构地区：[1]湖南大学信息科学与工程学院 [2]浙江猛凌机电科技股份有限公司 [3]中国人民解放军95899部队

出　　处：《电子测量与仪器学报》2014年第4期387-394,共8页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(61371115);湖南省高校创新平台基金(09K16)项目资助

摘　　要：针对药品生产企业安瓿瓶可见异物自动检测设备对安瓿瓶外观的要求,提出一种高效的安瓿瓶外观检测方法。该方法首先对所截取的瓶头数字图像进行中值滤波,然后利用Roberts交叉算子进行图像增强,通过选取适当的阈值将图像分割成仅由瓶头边缘和背景组成的二值化图像,最后利用梯度向量对边缘进行特征提取,通过分析中心和高度偏差特征来评价安瓿瓶瓶头的对称性和扁平度。实验结果表明,该方法检测正确率达到97%,能够有效检测出安瓿瓶常见的扁头、尖头、泡头、瘪头、断头等玻璃瑕疵,所筛选出的安瓿瓶能很好地满足安瓿瓶可见异物自动检测设备对瓶头外观的要求。According to the requirements of ampoule bottle appearance for automatic devices used to detect visible foreign matters in ampoule bottles of the drug production enterprises,a new method for ampoule bottle appearance detection will be presented in this paper,which implement such detection by extracting the features of ampoule bottle appearance based on gradient vectors.Firstly,the median filter operation should be applied to the gray images of the ampoule bottle head.Then the gray image will be transformed to a binary one which only includes the edge and its background of the bottle head by select appropriate threshold after image enhancement which fulfilled by the Roberts-Cross-Operator.Finally,the edge features of the bottle head are extracted by gradient vectors,sequentially the flatness and symmetry of the ampoule bottle head can be evaluated by analysis the center and height deviation.The result of experiments shows that the approach presented in this paper can achieve correct detection rate of 97% and can detect the glass defects effectively,such as flat head,sharp head,bubble head,sunken head,decapitated head and other issues.The good ampoule bottles selected by this way can perfectly meet the demands of the automation equipments for detecting visible foreign matter in ampoule bottles.

关 键 词：外观检测 图像增强 图像分割 特征提取 梯度向量 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]