In this study, we utilized Raman spectroscopy and deep learning to analyze wax impurities in honey filtration. We developed a custom siamese network with 1D convolutional and dense layers to compare Raman spectra of honey samples. The network, trained with a triplet loss, achieved 75% accuracy in distinguishing pure honey from wax-contaminated samples. We analyzed orange blossom honey in three conditions: pure, with wax, and wax-separated. The siamese network assessed impurities by comparing Raman spectra. In testing, the model effectively detected wax in new samples by measuring spectral distances. Additionally, a regression model provided a quantitative estimate of wax content with a very low test result (8.203648543319181e-25). This method enhances quality control in honey production, ensuring a purer final product.