Real-time moisture ratio study of drying date fruit chips based on on-line image attributes using kNN and random forest regression methods

Abstract

The purpose of this study is to eliminate weighing tools for moisture ratio (MR) estimation due to the high noise in the weighing system of Electro-Hydrodynamic (EHD) dryers. For this purpose, the product was photographed at specified time intervals during the drying process. Accordingly, the MR model of date fruit thin layers based on real-time color attributes and the environmental conditions of drying process was derived. Regression models were designed for each method using random forest (RF) and k-nearest neighbor (kNN) algorithms. Hyper-parameter tuning of RF and kNN models was done by random search and full search of parameters space, respectively. The variable space was also reduced in some of the models using these methods. This was done to reduce the output time in order to increase the detection speed. The MR was estimated by RF and kNN models with r2 of 0.976 and 0.959 in test set respectively. Since the accuracy of the predictive model was high, a correlation was found between the color change and the moisture content of date fruits. The overall results showed that by using online imaging, it is possible to accurately estimate the MR of the product in a convective EHD dryer without using any weighing system.