Determination of optimal harvest timing for field-grown apple fruits using hyperspectral imaging technology

Abstract

We utilized hyperspectral imaging technology, which is commonly used for nondestructive quality assessment in agriculture, to predict SSC (Brix, %) and also the firmness (N) of apples. In this research, various regression models were applied based on machine learning and deep learning with hyperspectral (400~1000 nm) spectrum data to predict SSC and firmness of apple fruits. To evaluate the prediction accuracy of each model, coefficient of determination (r square) and Root Mean Square Error (RMSE) was used. For this purpose, spectral data of apple fruits was acquired and prediction models using various regression models such as PLSR were developed. Also, various preprocessing methods were applied, including extracting meaningful pixels, MSC (Multiplicative Scatter Correction), SNV (Standard Normal Variate), to enhance the accuracy of regression models. Through these process, SSC and firmness prediction performance of each model was analyzed and compared with various combination of preprocessing methods.