基于近红外光谱的条斑紫菜菌落总数快速检测技术

为了探究快速、无损地检测条斑紫菜质量的可行性，本研究开发了一种基于近红外光谱技术的条斑紫菜微生物污染程度的定量分析方法。首先对来自不同海域的紫菜样本的菌落总数进行了测定，然后采集了155组样本的原始光谱信息和菌落总数信息。用标准正态变量变换（SNV）、多元散射校正（MSC）、二阶导数（Second-order derivative）等方法对光谱数据进行预处理。在完成最佳预处理方法筛选后，建立了基于光谱信息的非线性拟合（MLR）、支撑向量回归（SVR）、人工神经网络（ANN）、卷积神经网络（CNN）菌落总数预测模型。结果表明，标准正态变量变换与二阶导数的组合预处理效果最优，基于全波段下深度学习模型CNN预测效果最好（r值为0.940）。由此说明，CNN作为一种深度学习模型，可以实现针对条斑紫菜微生物品质的快速评价。

Rapid Detection of Total Bacterial Count of Porphyra yezoensis Based on Near Infrared Spectroscopy

To discuss the possibility of the non-destructive prediction of the total colonies of Porphyra yezoensis, this research explored a non-destructive method that using near infrared spectral imaging to predict the total colonies of Porphyra yezoensis. Porphyra yezoensis samples were measured the total colonies first, and then collected the original spectral information and total colonies of samples. Standard normal variable transformation (SNV), multiple scattering correction (MSC) and second order derivative were used to preprocess spectral data. After selecting the best pretreatment method in this study, the prediction models of total number of bacteria were established based on spectral information including mixed logistic regression (MLR), support vector regression (SVR), artificial neuro network (ANN) and convolutional neural networks (CNN). The results of the investigation showed that the second derivative method combined with standard normal variable transformation was the relative best pretreatment method. And the relative best prediction model was the CNN model which was based on the full-wave band, which the r value was 0.940. According to these results, the convolutional neural networks (CNN) could be used to predict the total number of colonies of Porphyra yezoensis.