Deep chemometrics with convolutional neural networks for the detection of honey adulteration using Fourier transform infrared spectroscopy.

Honey adulteration with low-cost syrups like glucose syrup (GS), rice syrup (RS), and high-fructose corn syrup (HFCS) remains a global concern. National laboratories require fast, non-destructive methods to ensure honey authenticity. This study employed Fourier Transform Infrared (FTIR) microscopy, spanning the mid-infrared and lower near-infrared regions, combined with chemometric and deep learning models to detect foreign sugars in honey. Three PLS-DA models were developed to identify RS, GS, and HFCS individually, demonstrating excellent performance on independent test sets: one model (10 latent variables) achieved perfect accuracy (1.00), precision (1.00/1.00), and F1-score (1.00/1.00); another with 8 (latent variables) attained accuracy of 0.97, balanced accuracy of 0.96, and F1-scores of 0.96/0.98; and a third (6 latent variables) obtained accuracy of 0.94, balanced accuracy of 0.95, and F1-scores of 0.94/0.95. Moreover, a unified CNN-ANN model was trained on a concatenated feature vector derived from multiple preprocessing treatments of the spectra: raw data, Standard Normal Variate (SNV), first derivative (Savitzky-Golay; window_length = 11, polyorder = 2), SNV + first derivative, second derivative (Savitzky-Golay; window_length = 11, polyorder = 3), and SNV + second derivative. After training for 50 epochs with Optuna-optimized hyperparameters, the model achieved a balanced accuracy of 0.79, precision of 0.81, sensitivity of 0.72, specificity of 0.85, and F1-score of 0.76 on the test set. This single classifier simplifies deployment by detecting all three adulterants simultaneously, eliminating the need for separate models. A user-friendly Python based interface was developed to enable batch or individual spectral analysis, automatically applying preprocessing and displaying results in a structured table with sample names, predicted classes (pure or adulterated), and visual flags. This integrated approach, combining FTIR spectroscopy, advanced modeling, and intuitive software, provides a rapid, reliable and practical solution for routine honey authenticity screening in quality control laboratories.