This project proposes an interferogram-based plastic classification method that, through augmentation techniques and a CNN model, achieves 97% accuracy with only 762 data points, demonstrating a more efficient and real-time alternative to spectrum-based approaches requiring larger datasets and longer training times.