Design and construction of real time sorting and grading machine of hazelnut using acoustic signal processing method

Naderi-Gar gari, Saber (2017) Design and construction of real time sorting and grading machine of hazelnut using acoustic signal processing method. Masters thesis, University of Mohaghegh Ardabili.

Preview

Text (طراحی و ساخت سیستم هوشمند آنلاین سورتینگ فندق با به‌کارگیری پردازش سیگنال آکوستیک و شبکه عصبی مصنوعی) Saber Naderi-Gar gari.pdf Download (580kB) | Preview

Abstract

: In this research, a method is presented for grading and sorting and categorizing of hazelnut by combining acoustic signal processing and artificial neural network (ANN) techniques. For this purpose a system was designed and constructed which acoustic signal was produced using a 130×130mm steel plate with thickness of 3mm. By contacting the hazelnut to the steel plate, the produced sound was measured by a microphone which was placed under the plate and it was transmitted to the computer by sound card to save in the computer memory and for the real time and next processes on MATLAB and LABVIEW2014 software. To avoid the excessive environment ambient noises a Infrared circuit and sensor was used. For designing the circuit and programming of Data acquisition Card, electronic software Express and LABVIEW were used, respectively. Time and wavelet domains were used in data acquisition. For sorting and grading of hazelnuts, MATLAB software was used... After extracting features of signals, The SOM Neural Network, which is a self-learning and validation tool, is selected. The choice of an optimal model based on the specificity of the individual Compression Cluster, which determines the number of neurons that are used concurrently For evaluating system performance, 400 hazelnuts were used in four groups including 100 sound large, 100 sound small, 100 defected, broken and100 sound empty nuts. The most suitable network of SOM for sorting hazelnut is for full large, full small, different dimension broken and empty hazelnuts As Structure 4 groups was obtained to have a 2000–1–4 architecture.. Finally the accuracy for full large, full small, different dimension broken and empty hazelnuts were 97.1%,98.2%, 94.1% and 89.7% respectively and it showed the total accuracy of 94.8% for system in offline mode. By using this neural Self-organizing network and Check online mode That compilation from MATLAB and LABVIEW software for100 hazelnuts for each group, the accuracy for full big, full small, different dimension broken and empty hazelnuts were (79.2%), (68.5%), (67.3%) and (69.4%), respectively and it showed the total accuracy of (71.1%)for system in online mode. it provides a high-level control, fast response rate software LABVIEW2014.

Actions (login required)

View Item