The growing volume of plastic waste has intensified the demand for efficient and automated methods to identify and sort polymeric materials, especially in recycling facilities where speed and accuracy are essential. Infrared spectroscopy is widely used for polymer identification due to its sensitivity to molecular vibrations. However, interpreting spectral data can be challenging when polymers have similar chemical structures, such as polypropylene and polyethylene. The objective of this work is to develop a methodology that employs the Gramian Angular Summation Field as a preprocessing step for a Convolutional Neural Network to distinguish polypropylene and polyethylene based on their infrared spectra. The dataset comprises 948 spectra of various post-consumer plastic materials. After converting the one-dimensional spectra into two-dimensional images, the neural network architecture includes a single convolutional layer, a dense layer, and a final output layer for multiclass classification, distinguishing between polyethylene, polypropylene, and other polymers. The model achieves average precision, recall, and accuracy of approximately 97% on the validation set and around 91% on the test set, indicating good generalization performance. By successfully integrating infrared spectroscopy with time-series-to-image conversion, this work shows that high-performance polymer classification is achievable with lightweight neural networks, reinforcing the potential of this approach for efficient and automated plastic sorting in future recycling systems.