Hyperspectral imaging and machine learning for the prediction of SSC in kiwi fruits


  • Jon Elias Moen
  • Vebjørn Nilsen
  • Katherine Brox Saidi
  • Eivind Kohmann
  • Binu Melit Devassy
  • Sony George


Solids Content (SSC) of the fruits in a non-destructive way. A database is created which includes the hyperspectral data acquired in the visible and near-infrared region (VNIR) and measurements done with a sugar meter.We have applied di?erent machine learning techniques to investigate the correlation between spectral information and the SSC. The models tested were support vector regression (SVR), k-nearest neighbor (KNN), partial least squares (PLS), and multiple linear regression (MLR) with di?erent variable selection techniques and dimensionality reduction. The best model at determining SSC was Uninformative Variable Elimination (UVE)-PLS, which had RMSE = 1.047 °Brix and R2 = 0.39 on the test set.