Publication: Buğday tohumluklarının sınıflandırılmasında yapay sinir ağları ve lojistik regresyon analizinin kullanılması
Abstract
Buğday danelerinin sınıflandırılması hem verim ve kalitenin değerlendirilmesi amacıyla ıslahçılar için hem de üretici ve tüketici açısından önemlidir. Çalışmada satışa hazır ortalama %11,75 nem oranına sahip, iki farklı ekmeklik buğday çeşidi kullanılmıştır. Görüntü işleme yöntemi yardımıyla elde edilen verilerden toplam 1868 daneye ait 13 fiziksel özellik belirlenmiştir. Belirlenen bu özelliklerden aralarında yüksek korelasyon katsayısına sahip olanlar belirlenerek, uygulanan modellere dahil edilmemiştir. Sınıflandırma modeli olarak, LR ve YSA modelleri uygulanmış, her iki modele ait bulgular karşılaştırılmıştır. LR modeli %83 genel sınıflandırma başarısına sahip olup; sınıflandırma başarısı sırasıyla Acar çeşidinde %82,5, Alada çeşidi % 83,5 olarak belirlenmiştir. Çeşitleri ayırmada en etkili ilk üç değişken ovallik, dikdörtgenlik ve en boy oranı olmuştur. İkinci model YSA için veri seti,%80 eğitim ve %20 test veri seti olarak ayrılmıştır. YSA 2 gizli katmanlı olup, birinci gizli katmanda 9, ikinci gizli katmanda 7 nöron bulundurmaktadır. Model çok katmanlı, ileri beslemeli geri yayılımlı bir ağ mimarisine sahiptir. Test veri setine ait genel doğru sınıflandırma oranı % 84,6’dır. Acar % 84,8 ve Alada % 84,6 doğru sınıflandırma oranına sahiptir. Modelde en etkili ilk üç değişken sırasıyla ovallik, maksimum yarıçap ve boydur.
Classification of wheat grains is significant for breeders using the purpose of evaluating yield and quality, as well as for producers and consumers. In the study, two different varieties of bread wheat with an average humidity of 11.75% were used. From the data obtained using the image processing method, 13 physical characteristics of a total of 1868 grains were determined. Criteria with a high correlation coefficient between these characteristics were determined and were not included in the applied models. As a classification model, LR and ANN models were applied, and the results of both models were compared. LR model has 83% general classification success; classification success was 82.5% in the Acar variety and 83.5% in the Alada variety, respectively. The first three most effective variables for separating varieties were ovallik, dikdörtgenlik, and aspect ratio. Data set for the second model ANN, 70% train and 30% is divided into test data set. ANN is 2 hidden layers, containing 9 neurons in the first hidden layer and 7 neurons in the second hidden layer. The Model has a multi-layer, forward-fed backward-propagated network architectureThe overall correct classification rate for the test dataset is 84.6%. Acar has a correct classification rate of 84.8% and Alada 84.6%. In the model, the first three most effective variables are ovallik, maximum radius, and lenght, respectively.
Classification of wheat grains is significant for breeders using the purpose of evaluating yield and quality, as well as for producers and consumers. In the study, two different varieties of bread wheat with an average humidity of 11.75% were used. From the data obtained using the image processing method, 13 physical characteristics of a total of 1868 grains were determined. Criteria with a high correlation coefficient between these characteristics were determined and were not included in the applied models. As a classification model, LR and ANN models were applied, and the results of both models were compared. LR model has 83% general classification success; classification success was 82.5% in the Acar variety and 83.5% in the Alada variety, respectively. The first three most effective variables for separating varieties were ovallik, dikdörtgenlik, and aspect ratio. Data set for the second model ANN, 70% train and 30% is divided into test data set. ANN is 2 hidden layers, containing 9 neurons in the first hidden layer and 7 neurons in the second hidden layer. The Model has a multi-layer, forward-fed backward-propagated network architectureThe overall correct classification rate for the test dataset is 84.6%. Acar has a correct classification rate of 84.8% and Alada 84.6%. In the model, the first three most effective variables are ovallik, maximum radius, and lenght, respectively.