Abstract:

In order to determine the variety from walnut leaves, each leaf must be examined in detail. Species that are very similar in color and shape to each other are very difficult to distinguish with the human eye. Examining and classifying plant leaves belonging to many classes one by one is not appropriate in terms of time and cost. Studies on walnut varieties in the literature are generally classified as a result of experimental studies in the laboratory environment. There are two or three different classes in studies using walnut leaf images. In this study, firstly, a unique walnut dataset obtained from 1751 walnut leaf images obtained from 18 different walnut varieties was created. Classification was made using deep learning methods on the original walnut dataset. It has been tested with CNN models, which are widely used in the literature, and some performance metrics are recorded and the results are compared. The images were first preprocessed for cropping, denoising and resizing. Classification was made using CNN models on the original dataset and augmented dataset with data augmentation method. It was seen that the VGG16 CNN model gave the best results both in the original dataset and the augmented dataset. In this model, the accucarcy result found with the original data set was 0.8552, while the accuracy result in the enhanced data set was 0.9055. When the accuracy values are examined, it is seen that walnut varieties are classified successfully.

Keywords: