基于THz-TDS技术的培氟沙星和氟罗沙星抗生素定性定量检测研究

作为两种常用的喹诺酮类抗生素，培氟沙星和氟罗沙星残留问题引起人们的高度关注，研发快速、高效的检测手段成为一种需求。采用太赫兹时域光谱(THz-TDS)技术对鱼粉基质中的培氟沙星、氟罗沙星进行了研究。制备培氟沙星、氟罗沙星、聚乙烯和鱼粉纯净物质以及培氟沙星-鱼粉和氟罗沙星-鱼粉17种不同浓度二元混合物的压片样品，共106个；对所有压片样品进行太赫兹光谱测量和分析；利用连续投影算法(SPA)结合支持向量机(SVM)和反向传播神经网络(BPNN)建立定性判别模型，对培氟沙星-鱼粉和氟罗沙星-鱼粉这两种混合物进行分类判别；利用特征频率处吸收系数建立偏最小二乘回归(PLSR)、BPNN、多元线性回归(MLR)定量预测模型，分别对两种混合物进行定量预测。结果表明：纯净培氟沙星在0.775和0.988 THz存在明显吸收峰，纯净氟罗沙星在0.919和1.088 THz存在明显吸收峰，聚乙烯对太赫兹波基本没有吸收，鱼粉无吸收峰，两种抗生素与鱼粉混合后的峰值出现在纯净抗生素的吸收峰附近；在定性判别中，SVM判别结果最佳，预测集判别准确率、精确率、召回率、F1得分分别为97.06%，97.22%，97.06%和97.06%；定量回归中，SPA-BPNN 模型用于预测培氟沙星-鱼粉结果最佳，预测集相关系数(R_p)、预测集均方根误差(RMSEP)分别为0.984 9和0.009 5，SPA-MLR模型用于预测氟罗沙星-鱼粉结果最佳，R_p和RMSEP分别为0.982 7和0.040 6。研究表明THz-TDS技术对鱼粉基质中培氟沙星、氟罗沙星进行定性定量检测是可行的，为畜禽行业中培氟沙星和氟罗沙星实际检测提供理论和技术参考。

Study on Qualitative and Quantitative Detection of Pefloxacin and Fleroxacin Veterinary Drugs Based on THz-TDS Technology

As two commonly used quinolone veterinary drugs, pefloxacin and fleroxacin residues have attracted great attention, and it is a demand to develop rapid and efficient detection methods. This paper uses Terahertz Time Domain Spectroscopy (THz-TDS) to study pefloxacin and fleroxacin in fish meal feeds matrix. Firstly, 106 compressed samples of pure substances of pefloxacin, fleroxacin, polyethylene and fish meal feed, and binary mixtures of pefloxacin-fish meal feed, and fleroxacin-fish meal feeds with 17 different concentrations were prepared. Secondly, terahertz spectrum measurement and analysis were carried out on all tableting samples. Then, the continuous projection algorithm (SPA) combined with a support vector machine (SVM) and backpropagation neural network (BPNN) were used to establish a qualitative discrimination model to classify and discriminate the mixture of pefloxacin-fish meal feeds and fleroxacin-fish meal feeds. At last, partial least square regression (PLSR), BPNN and multiple linear regression (MLR) quantitative prediction models were established using the absorption coefficient at the characteristic frequency. The results showed obvious absorption peaks of pure pefloxacin at 0.775 and 0.988 THz, and obvious absorption peaks of pure fleroxacin at 0.919 and 1.088 THz. Polyethylene had no absorption of THz wave, and fish meal feeds had no absorption peak. The absorption peaks of two mixtures appeared near the absorption peaks of pure antibiotics. In qualitative discrimination, SVM was the best model, and the accuracy, precision, recall and F1 scores of the prediction set were 97.06%, 97.22%, 97.06% and 97.14%, respectively. In the quantitative regression, SPA-BPNN was the best model for predicting pefloxacin-fish meal feeds, with correlation coefficient (R_p) and root mean square error (RMSEP) of prediction set being 0.984 9 and 0.009 5 respectively, and SPA-MLR was the best model for predicting fleroxacin-fish meal feeds, with R_p and RMSEP being 0.982 7 and 0.040 6 respectively. This study shows that THz-TDS technology is feasible for qualitative and quantitative detection of pefloxacin and fleroxacin in fish meal matrix, which provides theoretical and technical reference for practical detection of pefloxacin and fleroxacin in the livestock and poultry industry.