Document Type : Research Paper
Abstract
In this study, an intelligent system for sorting closed shell and open shell pistachio nuts, was designed and developed based on combined acoustic and artificial neural network (ANN) techniques. The system included a microphone, PC, material handling equipment and an air reject pneumatic mechanism. The microphone, placed under the steel plate, received sound signals generated by the pistachio nut impact and measured the features extracted from the sound signals of the closed and open shelled nuts. The system was evaluated using the Kaleghouchi variety of pistachio nut. Features necessary for identification were extracted from the analysis of the sound signals in the time and frequency domains by means of fast fourier transform (FFT), power spectral density (PSD) and principle component analysis (PCA). Finally, using PCA, seven features were isolated to separate the pistachio nuts, such as a 99.73% reduction in features. More than 40 different ANNs topologies, each having different numbers of neurons in their hidden layers, were designed and evaluated. The optimal model was selected after several evaluations that minimized mean square error (MSE) and correct separation rate (CSR). The optimal ANN model for this system was a 7-12-2 configuration. The total system accuracy (CSR) for the three pistachio split types (closed shell, open shell and thin split) were 96.7%, 97.3% and 93.1%, respectively.
Keywords
)