Akademik Veri Yönetim Sistemi
7th International Conference on Computational Mathematics and Engineering Sciences 20-21 May. 2023, Elazığ - Türkiye, Elazığ, Türkiye, 21 - 23 Mayıs 2023, cilt.1, sa.1, ss.71-88
Deep learning (DL) methods have become a popular approach in recent years to detect
and classify complex patterns in large data sets. The use of DL techniques in the food industry
is also increasing. These techniques are used in areas such as determination of food quality,
production efficiency, food diversity and disease prevention. Grape leaf classification is also a
very important problem in this field. Knowing the type of leaf is of great importance for vine
leaf producers. Since the leaves are similar to each other, the types are mixed with each other
and this causes the wrong product harvest. Accurate classification of grapevine leaves can
contribute to plant taxonomy and morphology. At the same time, the correct determination of
the grape species sheds light on plant phylogenetic studies. In this study, the correct
classification of vine leaves has been achieved by using deep learning methods. The dataset
consisting of 500 publicly available images was transformed into a dataset consisting of 2500
grape leaf images by applying different image enhancement techniques. Vine leaves were
classified using convolutional neural network (CNN) based deep learning architectures and
the accuracy performances of the models were evaluated. RestNet50, MobilNet, DenseNet
and VGG19 architectures were each trained for 160 and 80 epochs. Results were examined
based on criteria such as F1 score, accuracy, precision and recall. The highest accuracy rate
was found to be 96.60% using the ResNet50 architecture. The results show that all
architectures showed high performance in grape leaf classification. The application of DL in
agricultural image analysis has offered a new perspective for rapid and accurate identification
of grapevine species.