Classification of vine leaves using deep learning methods

Classification of vine leaves holds significant importance in the agriculture and food industries. Early detection of leaf diseases is critical for increasing productivity and maintaining product quality. Traditional methods can be inadequate in this process, whereas deep learning techniques offer effective solutions by providing high accuracy and speed. In this thesis, deep learning models were used to classify a total of 4000 vine leaf images obtained from 8 different vine species. The main objective of the research is to identify the model that can classify these leaf images with the highest accuracy and to develop a classification system that can be used in agricultural applications. Performance comparisons were made using ResNet50, ResNet101, ResNet152, MobileNet, Xception and DenseNet169 models. First, a dataset was created by processing vine leaf images through various preprocessing steps. Then, this dataset was trained and evaluated with each model separately. The performance of the models was analyzed considering criteria such as accuracy, training time, and model complexity. Experimental results showed that the ResNet101 model exhibited the highest performance with an accuracy rate of 88%, but it required a long training time of 17 hours. The Xception model demonstrated a performance close to ResNet101 with an accuracy rate of 86%, and it was considered preferable due to its notable processing time of 10 hours. The MobileNet model, with an accuracy rate of 80% and faster results (2 hours and 20 minutes), made it suitable for real-time applications. The DenseNet169 model, on the other hand, exhibited the lowest performance with an accuracy rate of 73%. These findings demonstrate that deep learning models can be effectively used for vine leaf classification, and selecting the right model is critical for both accuracy and efficiency. Future studies aim to increase the success rate with different neural network models in the literature and to improve image processing techniques. Additionally, based on the model with the highest accuracy, it is planned to develop a practical mobile application for agricultural producers and consumers. Keywords: Agricultural Applications, Deep Learning, DenseNet, MobileNet, ResNet, Vine Leaf Classification, Xception