Classification of common diseases in tomatoes using deep learning methods

Tomato is one of the most consumed vegetables in the world with a per capita consumption of approximately 20kg per year. One of the most important factors that negatively affect the yield and quality in tomato production is the diseases called harmful organisms. Tomato can be affected by many diseases during the planting and cultivation process depending on climatic and environmental factors. It is very important for producers to detect the diseases that may occur in tomato production and to take the necessary precautions. Today, artificial intelligence applications have become very popular. Artificial intelligence is used for the detection of plant diseases encountered in the field of agriculture. In this thesis, common diseases in tomato are classified by deep learning methods, which is a sub-branch of artificial intelligence. In order to collect samples for common diseases in tomato plants, field studies were carried out with tomato producers operating in Kirkuk province. After reaching a sufficient number of samples, meaningful images were obtained for the system using image processing techniques. The dataset created with these images was used to train a deep learning based system. The results obtained show that the developed deep learning system offers high accuracy and reliability in the classification of tomato diseases. In recent years, significant developments in artificial intelligence, especially in the field of deep learning, have been reflected in agricultural applications and have been widely used in critical processes such as disease detection. Tomato is an important topic in terms of disease detection, both because it is a widely consumed agricultural product worldwide and because of its susceptibility to many fungal, bacterial and viral diseases. In this study, a total of five classes of tomato diseases, namely tomato mildew, early leaf blight, lead mould, bacterial cancer and healthy tomatoes were classified. The data used is completely original and includes three different groups of leaf, red tomato and green tomato images taken in the production area. These data were augmented twice and a total of 6414 images were obtained. Eighty per cent of the data was used for training and 20 per cent for testing, and the results were evaluated with 21 different deep learning methods. The five most successful deep learning methods were reclassified with machine learning algorithms using five different feature extraction methods and five-fold cross-validation while maintaining the same training/testing ratios. As a result of this classification study, the most successful results were obtained by classifying the best 100 features selected by Chi² (Chi-square) method among 1000 features obtained with EfficientNet-B0 model with Fine KNN machine learning algorithm. With this method, 88.4% training accuracy and 92% test accuracy were achieved.