Novel Machine Learning Approaches for Accurate Leaf Area Estimation in Apples

Abstract

This study aims to extend the existing leaf area (LA) models in the literature by introducing four machine learning (ML) models: the extreme learning machine (ELM), k-nearest neighbor algorithm (KNN), random forest (RF), and multilayer perceptron (MLP). The leaf length (L) and width (W) measurements were based on the longest and largest parts of the apple leaf samples. Subsequently, the digital planimeter was utilized to determine the actual LA. Three statistical metrics, namely root mean square error (RMSE), the mean absolute error (MAE), and the coefficient of determination (R2), were used to evaluate and validate the predictive accuracies of models. The results revealed that during the testing phase, the RF model outperformed others in estimating LA in apples, showing an RMSE of 0.924 cm2, MAE of 0.579 cm2, and an R2 of 0.994. Meanwhile, the MLP model displayed slightly lower performance with an RMSE of 2.963 cm2, MAE of 1.866 cm2, and R2 of 0.944. Overall, RF, KNN, and ELM models have showed remarkable efficacy as ML techniques for predicting LA. Their demonstrated effectiveness suggests their potential contribution to future studies within this area.