近红外光谱的海水微塑料快速识别

Study on Rapid Recognition of Microplastics Based on Infrared Spectroscopy

光谱技术与机器学习算法结合快速识别微塑料,为微塑料的现场检测提供了极大的技术支持,是一个得到极大关注的新领域。近红外光谱检测技术具有检测速度快、灵敏度高、不损坏样品,且可以在不对样品进行预处理的情况下直接检测等特点,在化学分析、质量检测等领域广泛应用。本文基于近红外光谱检测技术,研究比较了结合Support Vector Machine(SVM)和Extreme Gradient Boosting(XGBoost)两种机器学习分类算法,构建微塑料的高速有效识别分类模型。采用微型近红外光谱仪采集了20种常见的微塑料标准样品的光谱数据,为了防止过拟合,对每种样品多次采样,共收集了1260个微塑料样本,每个样本包含512个数据点。利用XGBoost算法进行特征重要性排序,共提取了对识别准确率影响较大的65个数据点。分别采用SVM算法和XGBoost算法对数据降维后提取的65个数据点建立微塑料快速识别模型,并运用网格搜索(GridSearchCV)对XGBoost算法影响较大的超参数进行选取,确定n_estimators,learning_rate,min_child_weigh,max_depth,gamma的最佳超参数分别为700,0.07,1,1,0.0。为了提高模型的稳定性,识别速率和泛化能力,对模型采用10折交叉验证和混淆矩阵评估;研究结果表明,XGBoost模型对微塑料的识别准确率为97%,而SVM模型对微塑料的识别准确率为95%;XGBoost模型对微塑料识别的正确率优于SVM模型。综上所述,XGBoost模型微塑料识别整体性能优于SVM模型,为实际微塑料快速识别提供技术支撑。

The combination of spectroscopic technology and machine learning algorithm for rapid identification of microplastics provides great technical support for microplastics’ field detection, a new field that has attracted great attention. Nirs detection technology has the characteristics of fast detection speediness, highly sensitization, damage less, and can be directly detected without sample pretreatment, widely used in chemical analysis quality detection and other fields.This paper compares support vector machine(SVM) and Extreme Gradient Boosting(XGBoost), two machine learning classification algorithms based on the infrared spectrum, to build a classification model for high-speed and effective recognition of microplastics. Acrylonitrile butadiene styrene(ABS), Polyacrylonitrile(PAN), Polycarbonate(PC), Polyethylene glycol terephthalate(PET), Polymethyl methacrylate(PMMA), Polypropylene(PP), Polystyrene(PS), Polyvinyl chloride(PVC), Thermoplastic polyurethane(TPU), Ethylene-vinyl acetate copolymer(EVA), Polybutylene terephthalate(PBT), Polycaprolactone(PCL), Polyethersulfone(PES), Polylactic acid(PLA), Polyoxymethylene(POM), Polyphenylene Oxide(PPO), Polyphenylene sulfide(PPS), Poly tetra fluoroethylene(PTFE), polyvinyl alcohol(PVA), Styrenic Block Copolymers(SBS)20 standard samples of microplastics were collected by using A miniature near-infrared spectrum. In order to prevent overfitting, 1 260 microplastic samples were collected, each sample containing 512 data points. The XGBoost algorithm was used to rank the importance of the logarithmic data points, and a total of 65 data points which greatly influenced the recognition accuracy were extracted. SVM algorithm and XGBoost algorithm are respectively used to establish a microplastic fast recognition model for 65 data points extracted after dimensionality reduction, and GridSearchCV is used to select the hyperparameters that have a great influence on XGBoost algorithm to determine n_estimators, learning_rate, The optimal hyperparameters for min_child_weigh, max_depth, and gamma are 700, 0.07, 1,1, 0.0, respectively. In order to improve the model’s stability, recognition rate and generalization ability, a 10-fold cross-validation and confusion matrix were used to evaluate the model. The results show that the recognition accuracy of the XGBoost model is 97%, while that of the SVM model is 95%. The accuracy of the XGBoost model is better than the SVM model. In conclusion, the overall performance of the XGBoost model was better than that of the SVM model, which provides technical support for rapid identification of actual microplastics.