Neural network with k-fold cross validation for oil palm fruit ripeness prediction

Abstract

The combination of hyperspectral imaging and artificial neural network (ANN) can predict fruit ripeness. This work investigated the application of hyperspectral imaging and ANN models with the k-fold cross-validation method for ripeness prediction of oil palm fresh fruit bunches (FFB) for inline sorting and grading machine vision. Crude palm oil (CPO) is an exporting commodity for countries such as Indonesia and Malaysia. Oil palm FFB ripeness determines the quality of CPO. The unique shapes and colors of FFBs need innovative methods to substitute tedious and cumbersome manual sorting and grading. The oil palm FFB samples used in this study were categorized previously based on color and loosed fruits. We applied the Savitzky-Golay (SG) smoothing filter and 7-fold cross-validation for hyperspectral datasets before being used for the ANN models and a confusion matrix to find the ANN model accuracies. We obtained 72 data points after SG filter and data selection from 523 data points. The prediction results showed an average accuracy of 79.48%, in which three folds with k of 2, 5, and 7 gave the highest accuracy of 90%. The results confirmed the potential use of hyperspectral imaging, with k-fold cross-validation and ANN models for ripeness prediction of oil palm FFBs.