鸡蛋新鲜度是反映鸡蛋内部品质的一个重要指标。 为了能够实现鸡蛋新鲜度的快速无损检测, 利用微型光纤光谱仪采集鸡蛋550～950 nm的透射率光谱曲线, 与鸡蛋的哈夫单位值进行了定量分析。 通过不同的预处理方式分别结合偏最小二乘回归(partial least squares regression, PLSR)与支持向量回归(support vector regression, SVR)建立模型, 比较了不同模型的预测结果, 发现一阶微分结合SVR能够实现较好地预测, 且利用SVR建模要优于PLSR。 为了提高运算效率, 减少无用信息对建模的不良影响, 分别利用线性降维主成分分析法(principal component analysis, PCA)与非线性降维局部线性嵌入(locally linear embedding, LLE)对一阶微分后的光谱数据降维, 比较两种降维方法的预测效果, 得出了LLE降维要优于PCA降维, 其训练集和预测集的相关系数与均方根误差分别为92.2%, 7.21和91.1%, 8.80, 训练集交叉验证的均方根误差相比减少了0.79。 实验结果表明, 利用局部线性嵌入结合支持向量回归进行非线性建模, 能够提高鸡蛋新鲜度的预测能力, 表明该方法对鸡蛋新鲜度的可见/近红外光谱检测可行。

The freshness of egg is an important index to reflect the internal quality. In order to achieve non-destructive detection of freshness, micro fiber spectrometer was used to sample 550～950 nm transmittance spectra of eggs which performed quantitative analysis with haugh unit of eggs. Different pretreatment was combined with partial least squares regression(PLS) and support vector regression(SVR) respectively to find that first derivative combined with SVR predicted better than others through comparison, and it was better to model by SVR than by PLS. In order to improve efficiency and decrease adverse effects of useless information for modeling, the linear dimensionality reduction with principal component analysis (PCA) and the nonlinear dimensionality reduction with locally linear embedding(LLE) were used for the data of first derivative respectively. It indicated that LLE was better than PCA after comparison, and the correlation coefficient of calibration and prediction were 92.2%, 91.1%, and the root mean square error were 7.21, 8.80. The root mean square error of cross validation decreased 0.79.The experimental result illustrated that the nonlinear model of LLE combined with SVR improved predictive performance of egg freshness. It is feasible for the detection of visible/near-infrared spectrum of egg freshness to apply this method.