基于变量筛选的温州蜜桔品质的光谱快速检测

Rapid spectral detection of satsuma quality in wenzhou based on variable screening

中文关键词:  蜜桔  可见/近红外光谱  变量选择  可溶性固形物

英文关键词:satsuma  visible-near infrared spectroscopy  variable selection  soluble solids

中文摘要:

      目的 利用可见/近红外光谱技术结合变量筛选算法建立预测模型。方法 采集7个不同批次蜜桔样本的漫透射光谱, 预处理优化后, 以无信息变量消除法(uninformative variable elimination, UVE)、竞争性自适应重加权法(competitive adaptive reweighting sampling, CARS)及其组合(UVE-CARS)共3种策略来进行光谱有效波段的筛选, 建立蜜桔可溶性固形物含量(soluble solid content, SSC)的偏最小二乘预测模型(partial least square, PLS)。结果 比较全变量模型和3个特征变量模型的预测性能, UVE-CARS-PLS模型取得了最优的检测效果, 相比全变量模型, 建模变量数减少了96.5%,其预测集相关系数RP提升至0.732, 预测集均方根误差 (root-mean-square error, RMSEP)下降至0.873 0Brix。结论 结合多重变量选择算法, 可以进一步压缩建模变量数, 简化模型, 提高模型预测精度, 实现区域蜜桔品质的光谱快速检测。

英文摘要:

      Objective To establish a prediction model by using visible-near infrared spectroscopy technology and variable selection algorithms. Methods The diffused transmission spectra of seven different batches of satsumas were collected, and then the spectra were optimized using preprocess methods. Effective spectrum bands were screened by 3 strategies, including uninformative variable elimination (UVE), competitive adaptive reweighting sampling (CARS) and its combination (UVE-CARS), and partial least squares (PLS) prediction model for the soluble solids content (SSC) of satsuma was established Results Comparing the prediction performance of the full variable model and the 3 characteristic variable models, the UVE-CARS-PLS model achieved the best detection effect. Compared with the full variable model, the number of modeling variables was reduced by 96.5%, and the correlation coefficient of prediction set (RP) reached 0.732 and root mean square error (RMSEP) decreased to 0.873 0Brix. Conclusion Combined with the multiple variable selection algorithm, the number of modeling variables can be further compressed, the model can be simplified, the prediction accuracy of the model can be improved, and the spectral detection of regional tangerine quality can be achieved quickly.