Predicting Mechanical Properties of Golden Delicious Apple Using Ultrasound Technique and Artificial Neural Network

Abstract

Non-destructive quality assessment of Golden Delicious apple is the aim of this study. The ultrasound non-destructive testing (NDT) system in the range of 40 kHz–20 MHz ultrasonic wave was applied in combination with artificial neural network (ANN) as a powerful modeling tool. The studied quality factors were the mechanical properties including firmness, elastic modulus, and stiffness of apples. To develop the ANN models, the feed-forward neural network with the backpropagation algorithm was developed. For firmness, the best ANN model with 7-11-1 topology (7 features as inputs, 11 neurons in the hidden layer, and one output) had the highest R2Prediction and the lowest MAEPrediction and SEPrediction which were equal to 0.999, 0.0836 N, and 0.1119 N, respectively. For elastic modulus, the 7-17-1 structure of ANN was the best with corresponding R2Prediction equal to 0.999, MAEPrediction equal to 0.0087 MPa, and SEPrediction equal to 0.0117 MPa. The model with structure of 7-13-1 and R2Prediction of 0.999, MAEprediction of 0.027 N/mm, and SEPrediction of 0.0371 N/mm was also selected for predicting the stiffness of apple. The results proved the ability of this method to non-destructively predict mechanical properties of Golden Delicious apple.