基于近红外高光谱成像技术的鸡蛋污染过程中菌落总数可视化研究

Visualization of the total viable count of bacteria during the polluted process of eggs based on near infrared hyperspectral imaging technology

[目的]为探讨鸡蛋污染过程中微生物总量无损预测的可行性,本文研究了一种基于近红外高光谱成像技术可视化鸡蛋中菌落总数的无损预测方法。[方法]将鸡蛋样本在接种大肠杆菌和铜绿假单胞菌的混合菌悬液后储存,采集并统计不同污染程度鸡蛋样本的原始高光谱信息和菌落总数信息;在完成最佳预处理方法筛选后,结合连续投影算法(SPA)分别建立了基于全波段和特征波段光谱信息下的偏最小二乘法(PLS)和支持向量机(SVM)菌落总数预测模型;优选出相对最佳预测模型后进一步实现对鸡蛋内部污染程度的可视化研究。[结果]二阶导数预处理效果相对最佳,其中交叉验证集相关系数R_(CV)为0.88,交叉验证集均方根误差为0.82 lg(CFU·g^(-1));鸡蛋中菌落总数的相对最佳预测模型为基于特征波段下SVM模型,其中建模集相关系数R_C为0.88,预测集相关系数R_P为0.84,建模集和预测集均方根误差分别为0.86和0.97 lg(CFU·g^(-1));根据鸡蛋内部污染程度及光谱特性的差异,建立了鸡蛋内部污染程度伪彩色图像。[结论]近红外高光谱成像技术结合化学计量学及图像处理方法,可实现对鸡蛋内部菌落总数的无损预测及污染程度的可视化,该技术可以为鸡蛋的安全性在线检测提供参考。

[Objectives]To discuss the possibility of the non-destructive prediction of the total viable count of bacteria in the polluted eggs,this paper explored a non-destructive method that using near infrared hyperspectral imaging to predict and visualize the total viable count of bacteria in eggs.[Methods]Egg samples were inoculated with the mixture solution of Escherichia coli and Pseudomonas aeruginosa,then the original hyperspectral information and the total viable count of bacteria of eggs in different pollution levels during storage were collected.After selecting the best pretreatment method in this experiment,the predicting model by partial least squares(PLS)and support vector machine(SVM)model were developed based on full wavelengths and the feature wavelengths via the successive projections algorithm(SPA).Finally,the relative best prediction model of the viable count of bacteria was selected and the visualization study of the bacteria was realized in eggs.[Results]The second derivative method was the relative best pretreatment method,the correlation index of cross validation(R CV)was 0.88,and the root mean square error of cross validation was 0.82 lg(CFU·g^-1).The relative best prediction model was the SVM model which was based on the feature wavelengths,the correlation index of calibration(R C)and prediction(R P)was 0.88 and 0.84,and the root mean square error of calibration and prediction was 0.86 and 0.97 lg(CFU·g^-1),respectively.The visualized images of the polluted levels in eggs were built based on the difference of the pollution levels and spectral characteristics.[Conclusions]The near infrared hyperspectral imaging technology,and multivariate statistical and image processing methods can realize the prediction and visualization of the total viable count of bacteria in eggs,and this technology can also provide the foundation for on-line detection of the egg security.