基于CUVE-PLS-DA的鸡蛋新鲜度在线检测分级

On-Line Detection and Classification of Egg Freshness Based on Consensus Uninformative Variable EliminationPartial Least Squares-Discriminant Analysis(CUVE-PLS-DA)

针对目前鸡蛋新鲜度检测技术方法存在劳动强度大、检测精度低、分级效率不足等缺陷,本研究在4 800枚/h的禽蛋传输机上搭建了可见-近红外透射光谱（501-1 000 nm）在线检测装置,动态采集鸡蛋透射光谱数据,并建立光谱信息与鸡蛋哈夫值等级的偏最小二乘判别模型。采用3∶1原则对鸡蛋样本进行随机划分,其中校正集169个,验证集57个,通过比较多种光谱预处理方法以及两种特征波长选择方法,得出标准正态变换预处理方法和多模式共识方法能够有效地提高模型的正确率、运算效率和预测能力,优化模型后的校正集和验证集准确率分别为92.31%、91.23%。结果表明本实验建立的可见-近红外光谱透射光谱检测方法能够对鸡蛋的新鲜度进行无损、智能、在线检测分级。

Although there are many methods available to detect egg freshness at present, they have shortcomings including laboriousness, low precision and low classification efficiency. An on-line monitoring device based on visible/near infrared spectroscopy（501–1 000 nm） was fitted to the 4 800 eggs per hour egg transport machine for the purpose of dynamically collecting transmittance spectral data for eggs. The collected data were used to establish a partial least squares discriminant（PLS-DA） model for the Haugh unit value of eggs. A total of 226 egg samples were randomly divided into two set： calibration set（n = 169） and validation set（n = 57）. By compared different spectral pretreatments and two wavelength selection methods, it was found that standard normal variate（SNV） transformation and multi-pattern consensus method could effectively improve the accuracy, efficiency and predictive ability of the PLS-DA model. The final calibration and validation accuracy were 92.31% and 91.23%, respectively. This study showed that visible-near infared spectroscopy could be used as a real-time and non-destructive detection method to classify egg freshness.