Flexible plastic packaging (FPP) constitutes one of the largest post-consumer plastic streams processed in recycling facilities. To address the key challenges of its sorting and quality control, this study developed and tested a classification procedure based on hyperspectral imaging (HSI), combined with machine learning. The aim was to automatically detect contaminants (i.e., other polymers and materials) within a polypropylene (PP) stream of FPP waste (FPPW). Hyperspectral images of representative FPPW samples of PP and contaminants were acquired in the short-wave infrared range (SWIR: 1000-2500 nm) and preprocessed using different combinations of algorithms to emphasize their spectral characteristics. Principal component analysis (PCA) was applied as exploratory analysis of the spectral data followed by the application of a hierarchical classification model, based on partial least squares-discriminant analysis (Hi-PLS-DA), to differentiate between PP and other materials considered as contaminants, including polyethylene, polyester, polyethylene terephthalate, polystyrene, cellulose, polyurethane, aluminum and multilayer films. The results showed a classification accuracy of 87.5 %, with 147 out of 168 flakes correctly identified, as verified by Fourier transform-infrared (FT-IR) spectroscopy, demonstrating the model robust performance in distinguishing PP from other materials. Assuming all correctly identified particles are properly sorted, the model is predicted to achieve a Recovery of 98.2 % by weight for PP, indicating minimal material losses, with a Grade of 94.4 % by weight, representing a significant improvement compared to 77.2 % in the initial feed FPPW stream. This work demonstrated the effectiveness of HSI combined with Hi-PLS-DA in developing an automatic and efficient sorting and/or quality control process for FPPW, with minor classification errors occurring in filaments and multilayer flakes.