Rice Leaf Diseases Detection Using Transfer Learning Models

Abstract

The "Rice Leaf Diseases Detection Using Transfer Learning Models" project aims todevelop an efficient and accurate system for identifying and classifying various rice leaf diseases using deep learning techniques. The project utilizes a dataset of 6,000 rice leaf images categorized into four disease types: Bacterial Blight, Tungro, Blast, and BrownSpot. The proposed methodology incorporates transfer learning models, includingResNet152V2, Xception, VGG19, MobileNetV2, and DenseNet201, which are fine-tunedto improve accuracy and generalization. The models are trained on the preprocesseddataset, employing image augmentation techniques such as rotation, flipping, and scalingto enhance model performance and prevent overfitting. After training, the models areevaluated on a test set to assess their classification accuracy. The results showthat DenseNet201 achieved the highest accuracy of 99.78%, followed by ResNet152V2at 99.33%, MobileNetV2 at 99.17%, VGG19 at 98.67%, and Xception at 97.28%. Theseresults demonstrate the potential of transfer learning in achieving high accuracyfor disease detection in rice leaves. The project’s outcome offers an innovative solutionfor real-time disease monitoring, potentially aiding farmers in making informed decisions andenhancing crop health management. Furthermore, the model can be deployed on webor mobile platforms, providing farmers with an accessible, user-friendly tool for earlydisease detection, thus contributing to more efficient agricultural practices and improvedcrop yields.