基于高光谱成像的香肠菌落总数回归预测及数据可视化

The Prediction of the Total Viable Count on Sausage Based with Hyperspectral Imaging Technique and Data Visualization

摘要:

香肠的好坏有很多种评价指标，菌落总数（TVC）是其中的一种。高光谱成像技术已经成为一种快速、无损检测食品品质的有效方法。本文利用高光谱成像技术对香肠的菌落总数进行了定量分析，对数据进行了主成分分析（PCA），研究发现数据集中前四个主成分累计贡献率已达97.65%，已经可以反映出香肠所包含的绝大部分信息。对前四个主成分对应的优化区间采用高斯核函数的SVM回归模型进行预测，并为了提高回归预测模型的精确度，对模型的c，g参数，进行了遗传算法（GA）、网格搜索算法和粒子群算法（PSO）寻优对比，其中PSO寻优可使回归预测值和真实值的相关系数为0.9777，交互验证均方根误差为0.0823，能够准确快速的实现香肠菌落总数的预测。除此之外，利用python对回归预测的数据进行可视化，更加直观的显示菌落总数变化，且可以达到实时观看的效果。

Abstract:

There are a lot of evaluation standard of the quality for sausage, one of which is the total viable count(TVC). Hyperspectral imaging technique has become an effective method to detect food rapidly and nondestructively. In this paper, Hyperspectral imaging technique has carried on the quantitative analysis to the total viable count (TVC) on the sausage. The data sets of sausage were assessed using the PCA method, and then the study found that the contribution rate of the first four principal component reaches 97.65% which can reflect the most information of the sausage. The SVM regression model based on Gaussian kernel function and the optimal interval the first four principal components is used to forecast TVC. In order to improve the accuracy of the regression model, the genetic algorithm (GA), grid search algorithm and particle swarm optimization (PSO) are compared to get the c and g parameters of the model. The correlation coefficient of regression prediction value and real value is 0.9777, and the root mean square error of interactive verification of PSO is 0.0823, which can accurately and quickly predict the TVC. Besides, Use python to realize visualization of regression prediction data which can show the change of TVC more intuitively and can achieve real-time watching.