稀疏降噪自编码结合高斯过程的近红外光谱药品鉴别方法

发布时间：2018-02-21 14:52

本文关键词： 高斯过程自编码小波变换近红外光谱药品鉴别　出处：《光谱学与光谱分析》2017年08期 　论文类型：期刊论文

【摘要】：提出一种稀疏降噪自编码结合高斯过程的近红外光谱药品鉴别方法。首先对近红外光谱数据进行小波变换以消除基线漂移,然后用稀疏降噪自编码(SDAE)网络提取光谱特征并降维表示,最后采用高斯过程(GP)进行二分类,其中GP选用光谱混合(SM)核函数作为协方差函数,记此分类网络为wSDAGSM。自编码网络具有很强的模型表示能力,高斯过程分类器在处理小样本数据时具有优势。wSDAGSM网络通过稀疏降噪自编码学习得到维数更低但更有价值的特征来表示输入数据,同时将具有很好表达力的光谱混合核作为高斯过程的协方差函数,有利于更准确的光谱数据分类。以琥乙红霉素及其他药品的近红外光谱为实验数据,将该方法与经过墨西哥帽小波变换的BP神经网络(wBP)、支持向量机(wSVM),SDAE结合Logistic二分类(wSDAL)、SDAE结合采用平方指数(SE)协方差核的GP二分类(wSDAGSE),以及未采用小波变换的SDAGSM网络等方法进行对比。实验结果表明,对光谱数据进行墨西哥帽小波变换预处理能有效提升SDAGSM网络的分类准确率和稳定性。wSDAGSM方法无论从分类准确率还是分类结果稳定性方面,都优于其他分类器。
[Abstract]:A method of drug identification based on sparse noise reduction self-coding and Gao Si process is proposed. Firstly, wavelet transform is applied to the near infrared spectrum data to eliminate baseline drift. Then spectral features are extracted and dimensionally reduced by sparse noise reduction self-coding SDAE network. Finally, Gao Si process GP) is used for two classification, in which GP selects spectral hybrid SMN kernel function as covariance function. Remember that this classification network is wSDAGSM.Self-coding network has strong model representation ability, Gao Si process classifier has the advantage in dealing with small sample data. The WSDAGSM network obtains lower dimension but more valuable features to represent input data by sparse de-noising self-coding learning. At the same time, the spectral mixed nucleus with good expressiveness is used as the covariance function of Gao Si process, which is beneficial to more accurate spectral data classification. The near infrared spectra of Erythromycin and other drugs are used as experimental data. The method is combined with BP neural network with Mexican hat wavelet transform (BP neural network), support vector machine (SVM) and Logistic binary classification (wSDAL / SDAE), GP binary classification using squared index SE) covariance kernel, and SDAGSM network without wavelet transform, etc. The experimental results show that, Pretreatment of spectral data with Mexican hat wavelet transform can effectively improve the classification accuracy and stability of SDAGSM networks. The proposed method is superior to other classifiers in terms of classification accuracy and stability of classification results.
【作者单位】：北京邮电大学自动化学院;桂林电子科技大学计算机与信息安全学院;中国食品药品检定研究院;
【基金】：国家自然科学基金项目(21365008,61562013) 广西自然科学基金项目(2013GXNSFBA019279)资助
【分类号】：O657.33;TQ460.72

【相似文献】