分析测试学报2025，Vol.44Issue(8)：1557-1567,11.DOI:10.12452/j.fxcsxb.250331247
基于可解释深度学习及表面增强拉曼光谱的微塑料高效识别方法
Efficient Identification of Microplastics Based on Interpretable Deep Learning-Surface-enhanced Raman Scattering
摘要
Abstract
Microplastic(MPs)pollution has become a major challenge to the global environment.Traditional methods have many limitations in detection of microplastics,highlighting the urgent need for highly sensitive detection technology without complex preprocessing.In this study,a novel frame-work with surface-enhanced Raman scattering substrate capture,deep learning recognition,and gra-dient-weighted class activation mapping(Grad-CAM)interpretation was constructed to solve the prob-lem of MPs detection.The results show that the Au nanosponge substrate can effectively capture MPs.The data enhancement and preprocessing techniques can effectively improve the prediction ac-curacy of the model.In addition,the classification accuracy of the one-dimensional convolutional neural network(1D-CNN)-based multi-branch binary classification network can be up to 85%,which is significantly higher than that of the machine learning model and conventional 1D-CNN model.The Grad-CAM analysis effectively elucidates the model´s decision-making rationale and provides in-sights into the causes of misclassification.This method was effectively validated using real-world mixed microplastic samples.The substrates employed in this study are characterized by their wide-spread material availability,straightforward fabrication process,cost-effectiveness,and significant potential for practical applications.关键词
表面增强拉曼光谱/微塑料/卷积神经网络/深度学习/梯度加权类激活映射Key words
surface-enhanced Raman spectroscopy/microplastics/convolutional neural networks/deep learning/gradient-weighted class activation mapping分类
化学化工引用本文复制引用
张艺严,马静,孙振丽,杜晶晶..基于可解释深度学习及表面增强拉曼光谱的微塑料高效识别方法[J].分析测试学报,2025,44(8):1557-1567,11.基金项目
国家自然科学基金资助项目(U21A20290) （U21A20290）
