ANN predicting of walnut seedlessness based on ultrasonic properties

With increased expectations for agricultural and food products of high quality and safety standards, the need for accurate, fast, cheap and objective quality determination of these properties in food products continues to grow. An Artificial Neural Network (ANN) predicting system was developed to automatically predict and detect full and seedless walnuts based on ultrasonic properties. Ultrasonic properties of walnuts including amplitude and attenuation coefficient in transmission mode were measured. Ultrasonic sensors which worked at the frequency of 33 kHz were employed. Then, artificial neural network was used for modeling the relationships. Unless, there were no significant difference between physical dimensions, ultrasonic properties remarkably were different for full and seedless walnuts. ANN model could successfully describe the relationships between ultrasonic properties and weight of walnuts using back-propagation learning method and inspection data. The best trained ANN model produced satisfactory correlation between measured and predicted values (0.873) and mean square error (0.0119). The results show that this method is applicable as a simple method for categorizing walnuts in two groups of full and seedless.

ANN, Walnut, Ultrasonic properties, Seedlessness