Scholarlink

<p>Traditional regression methods in materials science often face limitations in modeling nonlinear relationships, managing large datasets, and maintaining computational efficiency. As an alternative, artificial neural networks (ANNs), a core component of deep learning (DL), have demonstrated effectiveness in capturing complex patterns due to their generalization capacity, noise tolerance, and robustness. In this study, a multilayer perceptron (MLP) architecture was evaluated to predict the fingerprint region of FTIR spectra (1500–400&#xa0;cm⁻<sup>1</sup>). Six polymers—PET, HDPE, PVC, LDPE, PP, and PS—were analyzed using the FTIR-PLASTIC-c4 dataset. Performance was assessed via the coefficient of determination (R<sup>2</sup>), with polystyrene (PS) achieving the highest predictive accuracy (87%), followed by HDPE (84%) and PET (81%). These results indicate that the model effectively captures spectral variability for these polymers. Conversely, polyvinyl chloride (PVC) exhibited the lowest performance (62%), suggesting that further model optimization or refined feature selection may be necessary to enhance predictions for certain polymer types.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimization of FTIR spectral prediction of polymers in the fingerprint region with a deep learning model

  • 期刊论文

  • 发表日期：

    2026年3月17日

查看全文
  • O. Villegas-Camacho,
  • R. Alejo-Eleuterio,
  • I. Gaytan-Aguilar,
  • S. Martínez-Gallegos,
  • M. del C. Hernández-Berriel

摘要

Traditional regression methods in materials science often face limitations in modeling nonlinear relationships, managing large datasets, and maintaining computational efficiency. As an alternative, artificial neural networks (ANNs), a core component of deep learning (DL), have demonstrated effectiveness in capturing complex patterns due to their generalization capacity, noise tolerance, and robustness. In this study, a multilayer perceptron (MLP) architecture was evaluated to predict the fingerprint region of FTIR spectra (1500–400 cm⁻1). Six polymers—PET, HDPE, PVC, LDPE, PP, and PS—were analyzed using the FTIR-PLASTIC-c4 dataset. Performance was assessed via the coefficient of determination (R2), with polystyrene (PS) achieving the highest predictive accuracy (87%), followed by HDPE (84%) and PET (81%). These results indicate that the model effectively captures spectral variability for these polymers. Conversely, polyvinyl chloride (PVC) exhibited the lowest performance (62%), suggesting that further model optimization or refined feature selection may be necessary to enhance predictions for certain polymer types.

Graphical abstract