贮藏大米中霉菌菌落总数的高光谱检测研究

Hyperspectral Detection of the Total Number of Mold Colonies in Stored Rice

[目的]为了快速、精确检测贮藏大米中的霉菌菌落总数,拟用高光谱图像技术实现无损检测。[方法]采用SG-SNV detrending的方式对原始光谱数据预处理,去除基线散射,平滑光谱曲线;然后分别利用SPA算法和CARS算法选取反映大米霉菌菌落总数特性的特征波长组合,最后采用SVR方法分别在全光谱数据和2种特征光谱数据的基础上建立预测模型,对比分析各SVR模型的预测效果。[结果]基于CARS特征选择的模型(CARS-SVR)预测效果优于基于全光谱数据的SVR模型和基于SPA特征选择的模型(SPA-SVR),其预测集决定系数(R^2p)为0.8759、均方根误差(RMSEP)为0.0835。由于CARS-SVR模型的预测效果尚未达到农产品检测的精度要求,故引入GWO算法对SVR模型中的参数(c和g)寻优,优化后模型(CARS-GWO-SVR)的训练集和测试集决定系数(R^2c和R^2 p)分别达0.9621和0.9511。[结论]利用高光谱技术对贮藏大米中的霉菌菌落总数实施无损检测具有可行性,可为其他农产品的霉菌检测提供参考依据。

[Objective]In order to quickly and accurately detect the total number of mold colonies in stored rice,hyperspectral image technology was proposed to achieve non-destructive detection.[Method]The original spectral data was preprocessed through the method of SG-SNV detrending for purpose of removing the baseline scattering and smoothing spectral curves.Then the combination of feature wavelengths reflecting the mold characteristics was selected by SPA and CARS,respectively.On the basis of feature data and full spectral data,the prediction models of mold colonies in stored rice samples were established through SVR method,and the prediction results of models would be compared and analyzed.[Result]The effect of CARS-SVR model was better than that of SVR model and SPA-SVR model.The R^ 2 p and RMSEP of prediction set was 0.8759 and 0.0835.In order to improve the accuracy of CARS-SVR model,GWO algorithm was adopted to optimize c and g parameters in model,after that,the R^2 c and R ^2 p were improved to 0.9621 and 0.9511.[Conclusion]Hyperspectral imaging technology is feasible for non-destructive detection of the total number of mold colonies in stored rice,providing a promising tool for mold detection of other agricultural products.