Rapid Identification of Automotive Window Film Brands via Hyperspectral Imaging with Two-Stage Feature Selection

Abstract

This study presents a hyperspectral imaging-based method for identifying automotive window film brands. Hyperspectral data in the visible-near infrared (VNIR) and near infrared (NIR) range were collected and analyzed using partial least squares discriminant analysis (PLS-DA) to establish correlations between spectral characteristics and brand categories. To improve model performance, a two-stage feature selection strategy was developed, involving initial screening through band-label correlation analysis followed by feature compression using LASSO regression to extract the most discriminative spectral information. The results show that 23 key features were successfully extracted from the original 896 spectral bands, achieving 100% classification accuracy. Further analysis revealed that the method maintained 100% accuracy across multiple classifiers using only five core spectral bands, demonstrating high robustness and practical utility. This approach provides a reliable, rapid, and non-destructive technical solution for the brand authentication of automotive window films, with potential applications in market regulation, product certification, and related fields.