NIR Spectroscopy Analysis on the Insulating Paper DP by Data Augmentation-Enhanced CNN

Near-infrared (NIR) spectroscopy has been widely applied to assess the degree of polymerization (DP) of insulating paper. However, traditional chemometric methods relying on dimensionality reduction struggle with nonlinear spectral data and generalization issues, limiting their practical applicability. To address these challenges, this paper proposes a novel deep learning framework that integrates multifactor data augmentation with convolutional neural networks (DA-CNN). The data augmentation strategy simulates real-world variations in spectra, enhancing the model’s robustness against noise and measurement inconsistencies. Meanwhile, CNNs effectively capture deep spectral features, improving the model’s ability to learn complex nonlinear relationships between NIR spectra and DP values. Experimental results demonstrate that DA-CNN achieves a minimum root mean square error of 45.8 and a relative error rate below 10%, significantly outperforming conventional methods. This study provides a more accurate and generalizable approach for NIRS-based DP evaluation, offering new insights into deep learning applications in insulation diagnostics.