基于XGBoost算法的鱼粉中微塑料的红外光谱识别研究

Study on Infrared Spectral Recognition of Microplastics in Fishmeal Based on XGBoost Algorithm

作为四大新兴污染物之一的“微塑料”带来的危害日益凸显,微塑料的检测识别是其污染评估和风险管理防控的关键。以鱼粉饲料中的微塑料(包括PA、PE、PET、PP、PS、PVC)作为研究对象,运用XGBoost算法分别研究构建了近红外光谱和红外光谱定性识别模型。采用GridSearchCV工具包研究优化XGBoost模型的主要超参数,近红外光谱模型的超参数优化结果为n_estimators:300,learning_rate:0.08,gamma:0,max_depth:4,min_child_weight:1;红外光谱的超参数优化结果为n_estimators:100,learning_rate:0.02,gamma:0.20,max_depth:4,min_child_weight:1。基于优化后的超参数构建的近红外定性识别模型平均精确率(Precision)为0.985,平均召回率(Recall)为0.977,平均F1值(F1 score)为0.978,相比于优化前模型效果分别提升了40.17%,51.00%,50.00%;红外定性识别模型平均精确率(Precision)、平均召回率(Recall)和平均F1值(F1 score)均为1.000,优化后的模型效果分别提升了20.67%,27.50%,26.33%。进一步与PLS-DA模型对比分析发现,红外光谱的XGBoost模型与PLS-DA模型效果基本一致,近红外光谱的XGBoost模型各参数(Accuracy,Precision,Recall,F1 score)效果均不同程度地优于PLS-DA模型。综上所述,运用XGBoost算法可以有效识别鱼粉中不同种类的微塑料,该研究为鱼粉饲料中微塑料的快速检测识别方法提供理论支持和技术支撑。

As one of the four emerging pollutants,the harm caused by“microplastics”has become increasingly prominent.The detection and identification of microplastics are the keys to pollution assessment and risk management prevention and control.This paper uses microplastics(including PA,PE,PET,PP,PS,and PVC)in fishmeal as the research objects.The XGBoost algorithm studies and constructs the qualitative recognition models of near-infrared and infrared spectroscopy.The XGBoost algorithm studies and constructs the qualitative recognition models of near-infrared and infrared spectroscopy.Optimising the main hyperparameters of the XGBoost model using the GridSearchCV toolkit.The hyperparameter optimization results of the near-infrared spectroscopy model were n_estimators:300,learning_rate:0.08,gamma:0,max_depth:4,min_child_weight:1.The hyperparameter optimization results of infrared spectroscopy are n_estimators:100,learning_rate:0.02,gamma:0.20,max_depth:4,and min_child_weight:1.The average Precision of the NIR qualitative recognition model constructed based on the optimized hyperparameters was 0.985,the average Recall was 0.977,and the average F1 score was 0.978,which improved by 40.17%,51.00%,and 50.00% compared with the model before optimization.The average precision,average recall,and average F1 scores of the infrared qualitative recognition model were all 1.000,and the optimized model effect improved by 20.67%,27.50%,and 26.33%,respectively.Further comparative analysis with the PLS-DA model shows that the XGBoost model of the infrared spectrum is the same as that of the PLS-DA model,and the effect of each parameter(Accuracy,Precision,Recall,F1 score)of the XGBoost model of the near-infrared spectrum is better than that of PLS-DA model to varying degrees.In summary,the XGBoost algorithm can effectively identify different types of microplastics in fishmeal.This study provides theoretical and technical support for rapidly detecting and identifying microplastics in fishmeal.