Fıstık sınıflandırma sistemi için Siirt fıstığı imgelerinden gürbüz özniteliklerin çıkarılması

Siirt pistachio is mostly consumed as fresh cookies in the market because it consists of more coarsegrained than other types of pistachio varieties and differs according to the aromas and taste. In order to gain customer satisfaction, Siirt pistachio having annual production capacity of 25 thousand tons should be subjected to the fully automatic sorting and grading processes based on the quality standards. Currently classification process is carried out via manual labor and mechanical tools. Man powered classification is a time consuming, boring and inefficient process. In addition to this, such a process carries several risks in terms of food hygiene. On the other hand, mechanical sorting systems (vibration based and pin pickers) cannot achieved desired classification accuracies anymore. Consequently a fully automated system based on machine vision is required. In this study it is aimed to develop various image processing methods for extracting robust features from the acquired images of Siirt pistachio samples for using pistachio sorting and grading system. Thus diameter ratio, pistachio area and edge length over area features are extracted from the acquired images. Extracted proposed features are provided as input parameters for offline classification. Pistachios are grouped into colossal, big, medium, small, shell and uncracked classes by domain expert based on Turkish quality standards. Dataset is constructed from the totally 600 pistachios (each group has 100 samples). Naïve Bayes, Artificial Neural Network (ANN) and Support Vector Machine (SVM) are used as classifiers. Classifiers are evaluated based on generalization accuracies and Type II error (false negative or severe error) criterion. ANN having 83.33% accuracy outperforms others. ANN and Naïve Bayes are also selected as winner classifiers based on severe error (Type II error or false negative) constraint. In addition to these, proposed pistachio sorting and grading systems is subjected to the real time performance testing with respect the computational cost criteria. Experiments revealed that considering three images, total processing speed of one pistachio sample is approximately 314ms which indicates that average processing speed of the proposed pistachio grading system is 3nuts per second. Another significant fact is related to the computational cost of the Canny edge detector algorithm. It holds about 230ms for three images and therefore Canny edge process is the most expensive part that should be addressed in the future. It should be noted that the processing performance of the proposed system is for the single channel. Therefore customers can purchase several channels in order to establish parallel processing mechanism according to their needs and factory processing potentials. Consequently, ANN and Naïve Bayes classifiers along with proposed robust features are recommended to be using in pistachio sorting and grading system.