Determination of some vineyards disease with deep learning techniques

Turkey, which has the most important vineyard areas in the world, is one of the countries where the most grapes are produced. Vineyard diseases are one of the most important reasons that negatively affect the yield in viticulture. In this study, some connective diseases were defined and classified using artificial intelligence approach, Faster R-CNN, SSD Multibox and originally newly developed deep learning model. These diseases are powdery mildew, downy mildew, dead arm disease, grapevine leaf roll-associated virus disease (GLRaV) and grapevine fan leaf nepovirus (GFLV) diseases that are common and cause economic problems. The proposed method is trained and tested using 11 000 images. As a result of the experimental evaluations, the general accuracy rates in the detection and classification of diseases were found to be 92% for Faster R-CNN, 92.21% for SSD Multibox and 96.95% with the developed model. The proposed approach gave better results than similar methods in the literature. Therefore, it has been concluded that the method can be used reliably in the detection and classification of some ligament diseases. The main contribution of this study to the literature is the creation of a large new data set for five different connective diseases consisting of fungal and viral diseases, and the development of a original new deep learning model for disease detection and classification.