Özet:

Tomato is one of the most important vegetable species produced and consumed mostly in the geography we live in and in many parts of the world. One of the most important factors that negatively affect yield and quality in tomato production is a disease called harmful organisms. Tomatoes can suffer from many diseases at every stage of the planting process, depending on environmental and climatic factors. The first thing to do in plant diseases is to correctly identify the disease and take the necessary precautions. In this study, a dataset with a total of 18.160 tomato leaf images consisting of 9 diseased and 1 healthy class was used (Kaggle, 2021). Experimental results were obtained on the publicly accessible Kaggle tomato leaf disease dataset. During the analysis, the data set was divided into 80% training and 20% test data. In the study, 3000 features were extracted from tomato leaf images by using Convolutional Neural Network (CNN) based DenseNet-201, ResNet-101, and ShuffleNet models. In this study, dimension reduction was made using Principal Component Analysis (PCA) to reduce the number of features. Support-Vector Machines (SVM) algorithm was used to classify diseases in tomato leaves. The trained network architectures were examined one by one. As a result of these examinations, the accuracy rates of the architectures were found to be 93.5%, 97.1%, 91.0%, 94.5%, 97.4%, and 96.6% for AlexNet, DenseNet-201, GoogleNet, MobileNet, ResNet-101, and ShuffleNet, respectively. After the analysis, size reduction was made by combining DenseNet-201, ResNet-101, and ShuffleNet pre-trained network architectures with high accuracy and using Principal Component Analysis (PCA). According to the classification analyses made in this study, the Cubic SVM classifier and the One-vs-All Multi-Class component method, which performed the best, achieved 99.2% accuracy. As a result of the literature review, it was concluded that this study showed an effective and high performance for tomato leaf disease detection.