| Summary: | Quality of persimmon cv 'Rojo Brillante' fruits can be affected by changes in texture
during postharvest storage if storage conditions are not appropriate, which can impact
consumer acceptance. Therefore, developing accurate tools to predict texture before marketing
is of great interest. Hyperspectral imaging is a non-destructive technology that has been proven
effective for predicting internal attributes. This study aimed to evaluate the usefulness of
hyperspectral images for predicting flesh firmness. A total of 3,340 persimmons were stored for
three months under different temperature conditions (0°C, 1°C, and 5°C) to induce varying
changes in flesh texture. Hyperspectral images were acquired at harvest and every month for
each storage condition. Flesh firmness was measured using a texturometer immediately after
image acquisition. Samples were clustered into three groups using k-Means based on their
firmness data and mean spectra was extracted from each persimmon. Samples were randomly
divided into training (70%) and test (30%) sets. Besides, wavelength reduction was performed using Partial Least Squares Discriminant Analysis (PLS) coefficients. The training data was used
to train PLS, Support Vector Machine, and Random Forest models using all spectra and selected
wavelengths. All models achieved high accuracies, indicating that even using a few
wavelengths, it is possible to accurately predict the flesh firmness of persimmon fruits.
|
|---|