Prediction of Paddy Moisture Content during Thin La yer Drying Using Machine Vision and Artificial Neural N etworks

Abstract

The goal of this study was to predict the moisture content of paddy using machine vision and artificial neural networks (ANNs). The g rains were dried as thin layer with air temperatures of 30, 40, 50, 60, 70, and 80°C and air velocities of 0.54, 1.18, 1.56, 2.48 and 3.27 ms -1 . Kinetics of L*a*b* were measured. The air temperature, air velocity, and L*a*b* values were used as ANN inputs. The results showed that with increase in drying time, L* decreased, but a* and b* increased. The effect of air temperature and air v elocity on the L*a*b* values were significant (P< 0.01) and not signific ant (P> 0.05), respectively. Changing of color values at 80°C was more than other temperatures. The optimized ANN topology was found as 5-7-1 with Logsig transfer fu nction in hidden layer and Tansig in output layer. Mean square error, coefficient of det ermination, and mean absolute error of the optimized ANN were 0.001, 0.9630, and 0.031, re spectively. Keywords: Back propagation neural network, Color features, Im age processing, Rice.