ارزيابي يك دستگاه هوشمند براي جداسازي پسته با استفاده از شبكه هاي عصبي مصنوعي (ANN) و تبديل موجك انعكاس صدا

Using Wavelet Transformation and Neural Network for Detecting Blank (Hollow) Pistachio Nuts

One of the factors which profoundly affect the grading of pistachio nuts is the presence of hollow nuts in a batch. The hollow nuts are formed due to improper environmental conditions, with the individual blank nuts lacking solid a kernel. Currently, mechanical as well as pneumatic devices are employed to separate the hollow nuts from the folly developed ones. In this research the acoustic signals of the nutsʹ impact onto a steel plate (from two elevations of 25 and 35 cm) were used for separating the hollow nuts from the folly developed ones. To perform the tests, pistachio nuts were placed on a conveyor belt with a separation distance of 1, 3 and 5 cm and allowed to fall onto the steel plate. The digitized sound signals were processed in wavelet domain to find out the suitable features for detecting the hollow nuts. The obtained features were used as inputs to 280 multiplayer neural networks (MLP). The results indicated that a 20-6-2 network was the most suitable network for the separation. The test results indicated that the neural network software was able to detect the hollow nuts with an accuracy of 98% while the folly developed ones with 94%. The average recognition accuracy was 96% for 25 cm drop elevation while 87% for 35 cm drop elevation. There was no difference observed in the accuracy of the results when the nuts were placed at 1.0, 3.0 and 5.0 cm apart on the conveyor belt.