Capacitance based Non-destructive technique for post-harvest sugarcane quality prediction

Post-harvest deterioration of sugarcane quality is one of the major issues that affect sugar production in sugar industries. There is a need for an effective technique that could quickly measure sugarcane quality without pre-processing the sugarcane stalks. This study was intended to create a capacitance-based setup capable of predicting °brix through nondestructive testing. The setup measures relative permittivity values which along with weight is used to predict the value of °brix in sugarcane stalks. A 5V sinusoidal wave of 5 different frequencies in the range 1 kHz - 1 MHz was swept across the parallel plate capacitor with sugarcane stalk samples as the dielectric medium. A total of three hundred stalks from four commercial varieties of sugarcane (CO 95020, CO 86010, CO 86032 and CO 06022) were used for the analysis. A predictive model for °brix was built using multiple linear regression, random forest and artificial neural networks. Feature selection revealed that relative permittivity at 1kHz frequency and weight were highly capable of predicting °brix. The best prediction was obtained from the random forest model with a coefficient of determination (Rv
²
) of 0.942.