中文摘要：

目前常用的盐酸左氧氟沙星注射液含量的测定方法是高效液相色谱法，但此法不能应用于在线分析。文章利用近红外光谱分析技术分别与偏最小二乘（PLS）以及人工神经网络（ANN）的方法相结合，对同一厂家的35个不同批号的针剂样品分别建立了定量校正模型，并对随机抽取的12个样品进行了预测。首先，利用PLS的方法建立模型，得出模型的决定系数（R^2）和预测集样本的标准偏差（RMSEP）分别为0．964和0．2428，同时利用小波变换技术对光谱变量进行了高效的压缩，并利用了前馈神经网络建立了盐酸左氧氟沙星针剂的定量分析模型，利用该模型所得的R^2和RMSEP分别为0．944和0．5722。文章详细比较了两种方法的建模过程，相关参数选取的优化方法，实验结果令人满意，从比较结果来看，PLS方法略优于ANN方法，可以快速准确的给出该针剂的含量，具有无损，简单，快捷的特点，为近红外光谱技术应用于针剂的定量检测提供了一个新的有效方法。

英文摘要：

In the present study, a total of 47 levofloxacin hydrochloride injection samples were detected by near infrared （NIR） spectroscopy, and 37 samples were randomly selected to establish the quantitative models by partial least squares （PLS） and artificial neural network （ANN） technology, while other 12 samples were used for prediction. On the one hand, the model was established by PLS, the coefficient of determination （R^2） of the prediction is 0. 964, and the root mean squared error of prediction （RMSEP） is 0. 242 8. On the other hand, after the spectrum variables were highly effectively compressed using the wavelet transformation technology, the quantitative analysis model of levofloxaein hydrochloride was established through the ANN technology. The R^2 and RMSEP of the model is 0. 944 and 0. 572 2, respectively. In this work, we have a detailed comparison between the two technologies in the progress of two quantitative models and optimizing correlative parameter, and finally we got a satisfied result. The simulation experiment indicated that the above PLS model is more steady and precise than ANN model, which can get hold of a rapid and nondestructive quantitative analysis result of the injection. Thus, the research can provide powerful scientific basis and technical support for further analysis of levofloxacin hydrochloride injection.