Deep Learning-Based Damage Assessment in Cherry Leaves

This study aims to utilize deep learning methods for detecting diseases in cherry leaves to enhance agricultural productivity. While the detection of leaf diseases is currently performed by expert personnel, there may be a shortage of such experts, and the process can be time-consuming. Therefore, the primary objective of this study is to use deep learning-based disease detection applications to increase cherry production and enable early disease diagnosis. Additionally, the study investigates the impact of datasets on performance using two different datasets - one existing (PlantVillage Dataset) and one created for the study (Kozlu Dataset). Furthermore, the study examines the impact of hybrid architectures, combining convolutional neural networks (CNNs) and recurrent neural networks (RNNs), in addition to transfer learning methods and classical CNNs. On the PlantVillage dataset, AlexNet, VGG-16, MobileNet-V2, Inception-V3, and CNN models were compared. Due to the low performance of AlexNet and the long training time of VGG-16, MobileNet-V2, Inception-V3, CNN, and two different CNN+RNN models were compared on the Kozlu dataset. According to the average results, the MobileNet-V2 model achieved the highest accuracy and F1-score in both datasets. The methods were observed to perform somewhat better on the PlantVillage dataset compared to the Kozlu dataset. Additionally, hybrid models (CNN+RNN) were found to achieve higher performance than the classical CNN model. These findings indicate promising outcomes for deep learning models in cherry leaf disease detection. The best results in the study were obtained by the MobileNet-V2 and the proposed CNN + LSTM models. In future studies, the reliability of this study can be increased by using more diverse datasets, and disease detection performance can be enhanced by using different deep learning methods, leading to reduced disease detection times.