Evaluation of artificial neural networks (ANNs) in predicting the effects of cleaning, moisture content, temperature and time on the physical and microbial characteristics of wheat

Evaluation of physical characteristics of wheat grain during transportation, separation and storage of this valuable product plays a critical role. In this study, in the first step, effects of cleaning, moisture content, storage temperature and time, on some physical (hectoliter, thousand seed weight, and bulk density) and microbial (total count of microorganisms and mold counts) properties of wheat grain variety n-80 was determined and then obtained data were simulated by two Artificial Neural Network models including multilayer perceptron and radial basis function with different threshold functions and the data predicted by the ANNs were compared with experimental data. Our results revealed that multilayer perceptron network model with one hidden layer, and a TanhAxon – TanhAxon function of stimulus for the physical characteristics, with a structure of (5-11-3) including 5 inputs, 11 neurons in the hidden layer, and 3 outputs, with epoch of 3000, and also for microbial characteristics, with a structure of (5-4-2) including 5 inputs, 4 neurons in the hidden layer, and 2 outputs, with epoch of 4000, were the best ANN models for data predicting compared with radial basis function networks. Determination coefficients (R2) of hectoliter, thousand seed weight, bulk density, total count of microorganisms and mold count in multilayer perceptron network were 0.95, 0.989, 0.908, 0.908, and 0.938, respectively. The hectoliter and bulk density decreased with increasing levels of moisture content, storage time and temperature while they were increased with the rise of cleaning level. Also, the total count of microorganisms and mold count increased with increasing levels of moisture content, storage temperature and time and decreased with increasing levels of cleaning.