An efficient and precise (micro)plastic identification method: feature infrared spectra extraction based on EIS-VIP-CARS and ANN modeling

• Published in: Environmental research Vol. 279; no. Pt 2; p. 121916
• Main Authors: Zhang, Shuangsheng, Qiang, Jing, Liu, Hanhu, Zhou, Junjie, Li, Jun, Chen, Jian, Ding, Qiang, Qian, Kuimei
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier Inc 15.08.2025
• Subjects: Algorithms; ANN; CARS; Environmental Monitoring - methods; Equal interval sampling; Feature spectra; Microplastics; Microplastics - analysis; Neural Networks, Computer; Spectrophotometry, Infrared - methods; VIP score; CARS; PP; PS; DR; FWPs; ML; HSI; PBT; EIS; PCA; POM; MPs; EVA; TSS; AVRC; PCL; Microplastics; WPs; ANN; CNN; PLA; FTS; POPs; LDA; SVM; MNN; FS; PPO; FT; RMSECV; ABS; PPS; SNN; FGs; FTIR; PLS; TPU; SNT; GA; UMAP; SBS; Equal interval sampling; PTFE; CL; kNN; RMSE; VIP score; FTTS; HLs; NPs; Feature spectra; PC; PMMA; PE; TPs; PVA; MCS; VIP; PVC; PET; PES
• ISSN: 0013-9351; 1096-0953; 1096-0953
• DOI: 10.1016/j.envres.2025.121916

Understanding microplastics' (MPs) ecological impact necessitates their precise identification. To address the issue of the competitive adaptive reweighted sampling (CARS) algorithm extracting numerous feature wavenumber points (FWPs) that often miss transmittance peaks (TPs), resulting in high computational load and low accuracy in artificial neural network (ANN) models, this study introduces a novel approach. Initially, the equal interval sampling (EIS) method is employed to capture the main information of the full spectra. Subsequently, the variable importance in projection (VIP) is innovatively integrated into the CARS to formulate the EIS-VIP-CARS method for extracting feature spectra (FS). Using 20 typical MPs as the subjects, this study compares the identification performance of ANN models using full-spectra, EIS, CARS, EIS-CARS, VIP-CARS, and EIS-VIP-CARS. The results show that VIP-CARS extracts 128 FWPs, a reduction of 49.41 % compared to CARS. Moreover, the distribution of these FWPs is more concentrated around the TPs and their vicinity. The accuracy of MPs by the ANN model based on VIP-CARS is generally higher than that of CARS. EIS-VIP-CARS extracts 55 FWPs, representing a reduction of 58.65 % and 57.03 % compared to EIS and VIP-CARS, respectively. The overall distribution of these points closely aligns with the distribution of functional groups. The ANN model based on EIS-VIP-CARS can achieve a similar accuracy for MPs as the model based on EIS, both greater than 99 %, demonstrating good generalization ability. The accuracies of the MNN and convolutional neural network (CNN) models are higher than those of the SNN model, but the modeling time is longer. The ANN model established using the EIS-VIP-CARS is an efficient and precise approach for the identification of MPs in infrared spectroscopy. This study provides technical references for the research on the environmental behavior of MPs and is also of significant importance for the classification and management of plastic waste. [Display omitted] •Developed an EIS-VIP-CARS-based feature spectral extraction method.•VIP replaces AVRC in CARS to facilitate the capture of transmittance peaks.•Optimized hyperparameters to enhance the generalization ability of the ANN models.•CNN and MNN exhibit higher accuracies for MPs compared to SNN but require longer modeling times.