This paper (purchase required) proposes a strategy to differentiate premium from blended honey based on the two-dimensional correlation spectroscopy (2D-COS) of Raman spectra combined with multiple deep learning techniques. A reference set of 700 Raman spectra of Manuka, acacia and multi-floral honeys were collected, and the corresponding synchronous, asynchronous and integrative correlation spectra were obtained. The t-distributed stochastic neighbour embedding (t-SNE) and partial least square regression (PLSR) were used to analyze one-dimensional spectra and 2D-COS image datasets of the same sample, demonstrating that 2D-COS can highlight the complex fingerprint features of samples and thus contribute to the spectral characterization. The authors report that a combination of synchronous 2D-COS and deep residual shrinkage networks (DRSN) achieved the best performance compared to the other models, with the root mean square errors of prediction (RMSEP) of 3.1166 for Manuka honey and 2.3188 for acacia honey, respectively. They propose that this would be sufficient to quantify the adulteration of Manuka honey with cheaper honeys using a technique that is rapid, relatively cheap and non-destructive.
Photo by Benyamin Bohlouli on Unsplash
Comments