Prediction and classification of soluble solid contents to determine the maturity level of watermelon using visible and shortwave near infrared spectroscopy

The present work investigated the potential application of a portable and low-cost spectroscopic technique to predict the soluble solid content (SSC) for determining the maturity level of watermelons. A total of 63 watermelon samples were used in the present work, representing three different maturity levels: unmatured, matured, and over-matured. Before spectral acquisition, each watermelon sample was cut into half, producing 126 fruit portions. Visible shortwave near infrared (VSNIR) spectrometer was used to record the spectral data from the skin surface of each portion. The SSC of each portion was measured using a digital refractometer. Partial least square (PLS) regression method was used to establish both calibration and prediction models to predict the SSC values from the watermelon samples. Support vector machine (SVM) classifier was used to categorise spectral data into the respective maturity levels. Results showed that the coefficient of determination (R2) values for calibration models of unmatured, matured, and over-matured were 0.65, 0.81, and 0.78, respectively. For the prediction model, the R2 values for unmatured, matured, and over-matured were 0.60, 0.74, and 0.76, respectively. The SVM yielded good classification accuracy of 85%. The present work demonstrated that the proposed spectroscopic method could be applied to predict and classify the maturity level of watermelons based on their skin condition.