Bilgisayarlı Görü Sistemi Kullanılarak Siirt Fıstığının Otomatik Olarak Sınıflandırılması
Yükleniyor...
Dosyalar
Tarih
2015-09-15
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Erişim Hakkı
info:eu-repo/semantics/openAccess
Özet
Today, machine vision systems are used successfully and effectively on many areas including production, medical, military, robotic and agriculture. Delivery of agricultural foods, scrutinized before packaging, boosts the competitiveness of the organizations in the context of the quality standards and non-destructive testing. Turkey is the third pistachio producer in the world. According to 2010 data, Siirt province is the third producer in the Turkey and supplied 14 % of the national pistachio production. Siirt pistachios are consumed as a fresh nut, desired more than Antep pistachios in the market since, Siirt pistachios have coarser grains, contain low oil levels and higher nutrients properties. Hence, economic income would be better. According to Tigris Reconstruction Agency (TRA), a project of a modern plant in Siirt province with the processing capacity of 15.000 tons Siirt pistachios annually has been supported by IPA (Instrument for Pre-accession Assistance). It is expected that this modern plant will be active in the next three years. During the processing stages of the pistachios, there are three significant steps. These are detection of un cracked kernels, separation of empty pistachios from filled ones and classification of the pistachios according to their size, shape, healthy, morphological and aesthetical properties. Current technologies carry out these duties in a mechanical and physical manner at the expense of deterioration in the quality of the nuts. For example, separation of the empty kernels is realized based on buoyancy of water principle. However, the moisture content of pistachios and mold growth is increased and lead to aflatoxin contamination. Additionally, grading process is currently accomplished via mechanical sieves while changing the pores of the sieves. So, there is no automatic system that can separate pistachios with foreign materials, including in shell nuts, shells or shells fragments, various types of defects, mold, insect damage, rancidity and decay. Another concern is about cracked/un cracked classification. Detection of cracks in Siirt pistachios are currently performed manually by visual inspection of the workers in a primitive manner. There may be some health problems when considering food hygiene and safety. On the other hand, classification of the huge amount of pistachio nuts in this way may not bring the desired results since this tedious process actually is time consuming, labor intensive in nature and it is not cost effective. Because the amount of non-cracked shells directly influence the customer satisfaction, establishing such a classifier system with high accuracy rate becomes indispensable. Recently, optical, mechanical, electrical and acoustic methods are used for classification of the agricultural products. In this regard, Pearson showed that cracked/un cracked pistachio nuts can be well separated according to impact acoustic sounds. Furthermore, Kalkan et.al. achieved 96.7 % accuracy rate for the classification of the cracked/ un cracked hazelnut kernels. Additionally, Omid et.al showed that cracked Iran pistachios can be separated at a rate of 97.5%. Almost all studies in the literature utilized impact acoustic signal for extracting a feature vector. Nevertheless, the main limitation of the impact acoustic based system is that, they are more easily affected from the environmental noises and is successful only for the specific size range of pistachios. In case of different size and moisture content pistachios, fault rate can also be increased. Besides, for multichannel design in mind, misclassification rate may also increases due to interfering effect of the parallel channels. In order to address this particular problem, image data also can be utilized along with impact acoustic data. In this study, we aim to increase the classification accuracy of the proposed system by two different source of information (sound and image). To keep the impact acoustic data immune to environmental factors, piezoelectric sensor or vibration sensor will be stacked to the steel plate which is used for impact of pistachios. Thus, impact acoustic vibrations can be converted into the electrical signals robust from environmental factors and noises. This enhancement also makes the multichannel processing of pistachio separation feasible. In addition to impact acoustic, classification can be performed based on both spatial and volumetric features. General procedure will be as follows. Raw pistachio nuts will be left from high place within a certain time interval as a free fall motion. Images of these nuts will be acquired by passing them through a tunnel that contains several cameras with different angle of views. Then, pistachio nuts will bump into the steel plate with piezoelectric sensor in order to generate impact acoustic data. Analyzed pistachios based on image and impact acoustic will be grouped into cracked and un cracked groups in general. Un cracked and cracked nuts will be subjected to the quality standards, based on impact acoustic data with filled and un-filled kernels and image data with foreign materials, unmarketable species, moldy, first-second-and third degree quality, respectively. Pistachio samples will be supplied by different pistachio processing organizations in Siirt. By making the train and test set larger than the previous studies, real time performance of the proposed system will be predicted more accurately. As it is known, even 1 % raise in classification performance for industrial applications lead to add value and increase the competitiveness of the organizations. As a result, this productivity enhancement provides positive contributions to our national economy.
Açıklama
Anahtar Kelimeler
Robotics, Machine Vision, Image Processing, Classification, Sorting Machine
