基于近红外光谱检测和平衡级联稀疏分类的药品鉴别方法

发布时间：2018-01-16 11:15

本文关键词：基于近红外光谱检测和平衡级联稀疏分类的药品鉴别方法　出处：《光谱学与光谱分析》2017年02期 　论文类型：期刊论文

【摘要】：近红外光谱检测和模式识别方法相结合,在药品的现场快速无损监督管理中有广阔的应用前景。传统的鉴别方法以最小化错误率为目标,往往忽略了样本数据的类别不平衡性,从而使得少数类样本被多数类样本淹没,降低少数类样本对分类器的影响,使分类结果更加倾向正确识别多数类样本,严重影响鉴别结果。针对药品光谱数据中真假药品类别不平衡问题进行研究,融合平衡级联和稀疏分类方法(SRC),提出一种级联的稀疏分类药品鉴别方法(BC-SRC)。文中在多数类样本中选取和少数类数目相同的样本作为训练样本,并在多数类样本中进行多次平行采样使得多数类样本被全部获得过(采样次数为多数类样本数与少数类样本数商的向上取整),最终得到测试样本的多组预测结果,根据得到的多组结果获得最终预测标签。将提出的方法在Matlab 2012a上进行仿真实验,通过三组样本集的实验证明该方法的有效性,实验结果表明该方法优于常用的偏最小二乘(PLS)、极限学习机(ELM)和BP神经网络分类法,特别是在解决类别不平衡问题时,当不平衡因子大于10时,BC-SRC算法分类相对于其他算法性能更好,且稳定性更高。
[Abstract]:The combination of near infrared spectrum detection and pattern recognition has a wide application prospect in the field fast and nondestructive supervision management of drugs. The traditional identification methods aim at minimizing the error rate. The class imbalance of sample data is often ignored, which makes a few class samples submerged by most class samples, reduces the influence of minority samples on classifier, and makes the classification result more inclined to identify most class samples correctly. The problem of unbalance of true and false drugs in spectrum data was studied, and the balanced cascade and sparse classification methods were fused. A concatenated sparse classification drug identification method (BC-SRC) is proposed in this paper. In this paper, the same number of samples as a few classes are selected as training samples. And many parallel samples are sampled in most classes so that most of the samples have been obtained (the sampling times are the upward rounding of the sample number of most classes and the quotient of few samples). Finally, the multi-group prediction results of test samples are obtained, and the final prediction labels are obtained according to the obtained multi-group results. The proposed method is simulated on Matlab 2012a. The experimental results show that the proposed method is superior to the conventional partial least squares (PLS), extreme learning machine (ELM) and BP neural network classification methods. Especially, when the unbalance factor is greater than 10:00, the classification of BC-SRC algorithm is better and more stable than other algorithms.
【作者单位】：桂林电子科技大学电子工程与自动化学院;北京邮电大学自动化学院;中国食品药品检定研究院;
【基金】：国家自然科学基金项目(61562013,21365008) 桂林电子科技大学研究生创新项目(YJCXS201558) 广西物联网技术及产业化推进协同创新中心项目(WLW20060610)资助
【分类号】：R927;TP391.4;O433
【正文快照】： 2.北京邮电大学自动化学院,北京1008763.中国食品药品检定研究院,北京100050引言近红外光谱(near infrared reflectance spectrum,NIRS)能够对样品进行现场快速无损检测,实现对未知样本的定性或定量分析,同时可以通过光纤实现远距离测量,这些特点使得近红外光谱分析技术在食品

【相似文献】