Fruit leaf disease detection using deep convolutional neural network.

It is important to establish an autonomous detection system that will enable rapid recognition of plant diseases in agricultural production and will accelerate decision-making in the use of chemical pest drugs. Many countries are investing in the research of autonomous systems and developing many R&D studies. In our country, efficiency and smart agricultural practices in the agricultural ecosystem have been started to be supported as part of the Agricultural Monitoring and Information System (TARBİL). In this thesis, from an open source, image flipping, gamma correction, adding noise, PCA color expansion, A new dataset called "New Plant disease", 38 types of diseased and healthy, of 14 different plant leaves, was created with six different data expansion techniques as rotation and scaling. A 11-Layer CNN architecture has been created for approximately 83.107 images in the dataset that detect plant leaf disease. With this new model, the performance of the plant leaf disease classification problem has been examined in two ways. First of all, the data set has 4 varieties of plant leaves with the highest number of data and examined the performance of the model. Secondly, the performance of the model has been reviewed for the entire dataset. The results obtained are given on the thesis. In addition, Transfer Learning architectures such as GoogLeNet, ResNet50, Vgg-19, Vgg-16, DenseNet have been used for the entire dataset and their performance has been reviewed.