Non-destructive Estimation of Chemical Properties in Braeburn Apple using Convolutional Neural Network

Determination of appropriate time to harvest apples prevents its waste and also affects the quality of the fruit. This time depends on the variety, climate during the growing season and also the purpose of harvesting the fruit. The aim of this study is to predict non-destructive chemical properties related to harvesting such as starch, soluble solids and acidity using spectral data and implementation of deep learning algorithm. First, images of Braeburn apples were taken by hyperspectral camera in 4 different stages of maturity. Next, the spectral information was extracted. Then the chemical properties of starch, titrtable acidity (TA) and total soluble solids (TSS) were measured using destructive methods in laboratory. Eventually, the prediction model was created by convolutional neural network (CNN). The results illustrated that the coefficient of determination and the mean squared error for the properties of starch, TSS and TA were 95.4%, 4.8, 91.6%, 0.284, 84.2% and 0.424, respectively.