Performance Investigation of Pre-Trained Convolutional Neural Networks in Olive Leaf Disease Classification

Olive is a very significant crop grown in specific regions of our country. According to the data of the Ministry of Customs and Trade, with the production of approximately 420 thousand tons of table olives in 2019, more than 14% of the total production in the world was made in Turkey. Therefore, early diagnosis and treatment of diseases in olive leaves can lead to increased production capacity. Today, as in many fields, deep learning algorithms are widely used for the diagnosis of plant diseases. In this study, the classification of olive leaf diseases was carried out with the frequently preferred pre-trained deep learning networks such as AlexNet, SqueezeNet, ShuffleNet, and GoogleNet. In the data set, performance results were obtained for both the raw data set and the augmented data set by applying the data augmentation process. The obtained results were evaluated with the performance criteria as accuracy, sensitivity, specificity, precision, and F1-Score. While the best performance improvement was obtained for the accuracy value of AlexNet with 7.56%, the lowest improvement rate was obtained from the specificity value of ShuffleNet with 0.63%.