Non-destructive estimation of leaf area of durian (Durio zibethinus) – An artificial neural network approach

Abstract

Non destructive estimation of leaf area is necessary for taking successive observations on the same leaf. Usually, a common method for developing leaf area estimation models was to execute regression analysis by taking leaf area as dependant variable and leaf parameters as independent variables. However, the regression analysis applicable only when there is a relation between dependent and independent variables. Therefore, the objective of this work was to develop an artificial neural network (ANN) model including leaf length and leaf width as inputs and to make comparison with regression models. Leaf area (LA) of Durain (Durio zibethinus) was estimated by ANN and regression models. Among the linear regression models (LRM), the equation LA=0.888 (L*W)−4.961 (R2=0.91, MSE=5.17) was found to be the best fitting for durian. In ANN modelling best fitting results were obtained with two input nodes (leaf length and leaf width), one hidden layer, and one output layer (leaf area) as 2-1-1. Based on the values of root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and coefficient determination (R2), ANN model was found to be more accurate both in terms of training (R2=0.96, RMSE=3.77, MAE=2.41 and MAPE=7.77) and testing phases (R2=0.94, RMSE=4.81, MAE=2.91 and MAPE=0.05) than the best fitting regression model. Hence ANN method may be used as an alternative or supportive model in the estimation of leaf area of durian.