Transfer Learning for Rice Leaf Disease Detection

To feed the world’s population of 7.9 billion people, preventing crop failure through early disease detection is essential. Various bacterial, viral, or fungal diseases affect the rice leaf and these diseases drastically lower rice yield. Therefore, identifying rice leaf diseases is essential to meeting the demand for rice from an extensive worldwide population. However, the ability to identify rice leaf disease is constrained by the image backgrounds and the circumstances under which the images were captured. When tested on independent rice leaf diseased data, the performance of deep learning models for automated detection of rice leaf diseases suffers substantially. This stusy examines the results of well-known and widely used transfer learning models to detect the rice leaf disease. This can be done in two ways: frozen layers and fine-tuning. It was observed that the results of the freeze layers, the DenseNet169, achieved a good testing accuracy of 99.66%, and when the results of the fine-tuned transfer learning models were examined, Xception performed well and achieved 99.99% of testing accuracy.