This work aims to classify seven common household plastic types which include polyethylene terephthalate (PET), high density polyethylene (HDPE), polyvinyl chloride (PVC), low density polyethylene (LDPE), polypropylene (PP), polystyrene (PS), and polycarbonate (PC) utilizing near-infrared (NIR) spectroscopy. Four methods including linear discriminant analysis (LDA), partial least-squares discriminant analysis (PLS-DA), spectral angle mapper (SAM), and support vector machine (SVM) were tested for their classification performances, and principal component analysis (PCA) was applied before LDA and SVM. All the classification models were built based on virgin plastics. The results showed that seven types of plastic could be classified excellently with all the methods when the test sets were composed of virgin samples. When the models were tested on waste plastics, most types could be well classified, and all the misclassifications occurred between HDPE and LDPE and PET and PC. Then for HDPE and LDPE and PET and PC that were prone to be misidentified, some specific spectral bands were reselected for further classification. To achieve the best result, an approach combining PCA, SVM, LDA, and PLS-DA was presented. The validation results showed significant improvement, with the F1 scores of LDPE and HDPE increasing from 65.2% to 86.7% and 24.2% to 84.7%, respectively, and 100% accuracy was achieved for the other five types.

中文翻译：

---

基于近红外光谱的多种方法相结合的常见家庭塑料垃圾分类

本工作旨在对七种常见的家用塑料进行分类，包括聚对苯二甲酸乙二醇酯 (PET)、高密度聚乙烯 (HDPE)、聚氯乙烯 (PVC)、低密度聚乙烯 (LDPE)、聚丙烯 (PP)、聚苯乙烯 (PS) 和聚碳酸酯。 (PC) 利用近红外 (NIR) 光谱。测试了包括线性判别分析 (LDA)、偏最小二乘判别分析 (PLS-DA)、光谱角度映射器 (SAM) 和支持向量机 (SVM) 四种方法的分类性能和主成分分析 (PCA)在 LDA 和 SVM 之前应用。所有分类模型均基于原始塑料构建。结果表明，当测试集由原始样品组成时，所有方法都可以很好地对七种塑料进行分类。当模型在废塑料上进行测试时，大多数类型都可以很好地分类，并且所有错误分类都发生在 HDPE 和 LDPE 以及 PET 和 PC 之间。然后对于容易被误识别的HDPE和LDPE以及PET和PC，重新选择一些特定的光谱带进行进一步分类。为了获得最佳结果，提出了一种结合 PCA、SVM、LDA 和 PLS-DA 的方法。验证结果显示显着改善，LDPE和HDPE的F1分数分别从65.2%提高到86.7%和24.2%到84.7%，其他五种类型的准确率均达到100%。重新选择了一些特定的光谱带以进行进一步分类。为了获得最佳结果，提出了一种结合 PCA、SVM、LDA 和 PLS-DA 的方法。验证结果显示显着改善，LDPE和HDPE的F1分数分别从65.2%提高到86.7%和24.2%到84.7%，其他五种类型的准确率均达到100%。重新选择了一些特定的光谱带以进行进一步分类。为了获得最佳结果，提出了一种结合 PCA、SVM、LDA 和 PLS-DA 的方法。验证结果显示显着改善，LDPE和HDPE的F1分数分别从65.2%提高到86.7%和24.2%到84.7%，其他五种类型的准确率均达到100%。