Abstract:
Sweet corn (cultivar Incee2) is an important crop in Thailand. It can be processed to many kinds of products. The current problem is the method for sorting under an exacting standard. This experiment aimed to use the external features and chemical equations to predict the total soluble solid (TSS), the percentage of moisture and Texture. By considering electrical properties at various frequencies (0.012, 0.05, 0.1, 0.5, 1, 5, 10, 50, 100 and 200 KHz), it such as capacitance, inductance, impedance, resistance primary, resistance secondary, dissipation factor, quality factor, phase angle and the physical properties such as density, chroma, GMD and absorbance by photo sensor of sweet corn. Partial least squares regression (PLSR) was used to develop the calibration models. Samples were measured after harvest in every 6 hour. Independent variables were considered for establishment the models. The equation for total soluble solid (TSS) and texture of sweet corn with husk obtained R2 = 0.868, 0.750 by cross-validation, respectively. The equation for the total soluble solid (TSS), texture, moisture content (MC) of sweet corn without husk obtained R2 = 0.972, 0.961 and 0.530 by cross-validation, respectively. The results showed that the qualities of the sweet corn could be predicted by using the models from multivariate technique. Prediction of fresh or not fresh was considering electrical properties and physical properties. The partial least squares discriminant analysis (PLS-DA) was used to classify the freshness (0 = fresh, after harvesting time was less or equal than 24 hours and 1 = not fresh, after harvesting time was more than 24 hours). The classification models were developed and tested by cross validation. The results shows accuracy of classification for fresh sweet corn with husk obtained 100% and for unfresh sweet corn with husk obtained 68.75%. Total accuracy was 89.58%. The accuracy of classification for fresh sweet corn without husk obtained 90.63% and for unfresh sweet corn without husk obtained 87.5%. Total accuracy was 89.58%.