基于机器视觉的注射液针头胶帽缺陷检测

目的 封装检测是保证流水线上产品质量的关键。针对预罐装注射液，提出一种基于机器视觉的注射液针头胶帽缺陷的实用检测算法。方法 算法首先采用以灰度分布概率作为度量标准的直方图双峰法，对图像进行阈值分割；随后依据以特征角点为中心延伸出的4个象限区域进行特征分析，定位胶帽下边沿左、右角点，以计算胶帽高度；将对称轴点集进行分段直线拟合，得到对称轴所有可能的斜率和截距，基于边缘信息计算最优对称轴和胶帽倾斜角。结果 采用多组图像检验算法缺陷检测，实验结果显示检测成功率达到97.86%。结论 该算法能够对针头胶帽的多种缺陷进行检测，对不合格产品进行分类。

Defect Detection for Needle Cap of Injection Based on Machine Vision

Packaging detection is the key to ensuring product quality on the production line. The work aims to propose a practical algorithm based on machine vision for defect detection of the needle cap of the injection, regarding the prefilled injection. Firstly, the algorithm used the histogram bimodal method which took grayscale distribution probability as the measurement standard for the threshold segmentation of the image. Then, four regions extended with the characteristic corner as the center were subject to characteristic analysis. The detection of left and right corners at the lower edge of cap were made to calculate the cap height. The straight-line fitting of symmetry axis point set was conducted by segments to get all possible slopes and intercepts of the symmetry axis. The optimal symmetry axis and inclination of cap were calculated based on the edge information. The detection of algorithm defects was verified by multiple sets of images. The experimental results showed that the detection rate reached 97.86%. The proposed algorithm can detect various defects of the needle cap and classify the unqualified products.