基于集成分类器的泡罩包装药品缺陷识别

Defect Identification of Blister Packaging Medicine Based on Integrated Classifier

目的针对药品生产包装过程中常出现缺陷泡罩包装药品的问题,研究一种基于多特征构建与集成分类器的泡罩包装药品缺陷识别方法。方法该方法通过集成2个不同的分类器算法分别对药品图像类别进行预测,并采用联合判定函数对2个预测输出值进行联合决策,得到最终分类结果。第1个分类器模型通过将图像转化到HSV颜色空间,分割出泡罩区域和药片区域,进行特征设计,并在提取多项特征参数后构建BP神经网络分类算法给定药品类别预测。第2个分类器模型应用多层卷积神经网络取代传统算法对图像特征进行提取,并输出药品图像类别的预测值。根据2个分类器的性能进行算法集成,构成最终集成分类器。结果实验结果表明,该集成分类模型对数据集中泡罩包装药品图像进行分类识别测试,准确率达97%以上。结论集成分类模型不仅提高了单一分类器的识别准确率,也具有更佳的稳定性。该方法取得了卓越的分类效果,具有较高应用性。

The work aims to study a method of defect identification for blister packaging medicine based on multi-feature construction and integrated classifier in view of the problem that many defective blister packaging medicine products appear in pharmaceutical production process. Two different classifier algorithms were integrated to predict the categories of medicine image respectively, and the combined judgment function was designed to make joint decision on the two predicted output value to obtain the final classification result. In the first classifier model, the image was transformed into HSV color space to segment the blister region and pill region, and feature engineering was carried out and multiple feature parameters were extracted to construct BP neural network classification algorithm to predict the given medicine categories. In the second classifier model, multi-layer convolutional neural network was used to extract the image features instead of the traditional algorithm and output the prediction of medicine image categories. According to the performance of the two classifiers, the algorithm was integrated to the final integrated classifier. The experimental result showed that the data set was tested in classification and identification with this integrated classification model and the accuracy rate was more than 97%. The integrated classification model not only improves the identification accuracy of a single classifier, but also has better stability. This method has achieved prominent classification effect and has high applicability.