Classification of Hard Candies with Different Formulas by Using Hyperspectral Imaging Combining Chemometrics

Creative Commons License

Gülenç B. E., Eriklioğlu H., Çifte N. E., Doğdu S., Manley M., Öztop H. M.

22nd International Conference on Near Infrared Spectroscopy (NIR 2025), Rome, İtalya, 8 - 12 Haziran 2025, ss.21, (Özet Bildiri)

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: Rome
  • Basıldığı Ülke: İtalya
  • Sayfa Sayıları: ss.21
  • Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

Confectionary industry focuses on hard candy production and their quality parameters. Citric acid plays a key role in controlling the acidic structure and flavor profile of the candy. Isomalt, helps prevent sugar crystallization, giving the product a clear and smooth texture. Additionally, the use of sugar alcohols offers lower calorie options when replacing sucrose and is beneficial for sugar-free candies. The objective of this study is to explore Near Infrared-Hyperspectral Imaging (NIR-HSI) techniques, combined with chemometric modeling, can effectively predict hard candy samples in terms of citric acid concentration (1.5%, 2%, 2.5%), isomalt concentration (92.5%, 95%, 97.5%) and type of sugar alcohol (sorbitol, xylitol, erythritol) and enhance the quality of hard candy. NIR-HSI (Hyspex VNIR-1800) was employed in spectral range of 953-2500 nm. The analysis includes preprocessing steps like wavelength selection, Savitzky-Golay filter (15 windows, 2nd derivative), and normalization to improve data accuracy. In the modeling phase, k-Nearest Neighbors (kNN), Neural Networks, and Random Forest were employed to perform classification. Test set was randomly chosen to check the performance of each models on train set (3:7, test:train). Neural Network gave the highest average classification accuracy for citric acid concentration, isomalt concentration and type of sugar alcohol 95%, 99% and 97% respectively for test set. On the other hand, kNN algorithm had the lowest average classification accuracy with 95%, 92% and 90%. Results show that HSI can be utilized as a standard analysis method in hard candy production to not only determine the optimal formulation but also ensure the accurate identification and enhancement of quality parameters.

Keywords: Hard Candy, NIR-Hyperspectral Imaging, Chemometrics, Quality Control.