中文摘要：

为实现鸭蛋蛋清中庆大霉素（GM）残留含量的快速测定与检测模型精度的提高，应用遗传算法（GA）筛选导数同步荧光光谱特征波长，用遗传一支持向量回归（GA—SVR）建立鸭蛋蛋清中GM残留含量的预测模型。首先分析了样本的三维同步荧光光谱和确定了本实验研究的波长差△λ为1：70nm；然后利用sym5小波的2层分解对一阶导数同步荧光光谱进行去噪处理，并利用GA筛选出了14个荧光特征波长；最后利用GA优化了SVR的径向基核函数（RBF）参数（c，g，p），进而比较了GA—SVR、PLS和MLR3种预测模型的预测能力，研究表明，以GA—SVR模型的预测能力最强，其预测集的决定系数（R2）和均方根误差（RMSEP）分别为0．9830和1．1494mg／L。实验结果表明，GA能有效筛选出鸭蛋蛋清中GM的荧光特征波长和提高GA—SVR模型预测精度。

英文摘要：

To achieve the rapid determination of gentamicin （GM） in duck egg white and improve the accuracy of prediction model, the optimized characteristic spectral wavelengths are extracted from derivative synchronous fluorescence spectrum using genetic algorithm （GA）, and the prediction model of GM contents in duck egg white is developed by using GA-support vector regression （GA-SVR）. Firstly, 3-D synchronous fluorescence spectra of samples are analyzed and 120 nm is selected as the optimum wavelength difference in this paper. Secondly, the noise of the first order derivative synchronous spectrum is reduced by using the sym5 wavelet with 2 decompositions, and 14 characteristic wavelengths are selected by using GA. Lastly, the parameters （c, g, p） of RBF kernel function are optimized by using GA. Furthermore,the performance of 3 models of GA-SVR,PLS and MLR is compared,and the best prediction results are obtained by using the GA-SVR model. The experimental results show that the determination coefficient （R2） and the root mean squared error of prediction （RMSEP） for the GA-SVR model are 0. 983 0 and 1. 149 4 mg/L, respectively. This work proves that GA could effectively extract characteristic spectral wavelengths to determine the gentamicin content in duck egg white and improve the prediction accuracy of the GA-SVR model.