Tea Leaf Guard: An Enhanced Mobile App for Tea Leaf Disease Detection Using Ensemble Deep Learning Models

Our paper introduces Tea Leaf Guard, an ensemblebased transfer learning technique used to detect and classify the tea leaf diseases. The proposed approach focuses on the critical issue of the manual detection of tea leaf diseases. The research introduces a new ensemble model that combines the strengths of three state-of-the-art convolutional neural networks namely MobileNet, DenseNet, and Xception, via an optimized weighted voting mechanism. This model has been trained and tested on 886 images of eight different classes, which include seven diseased and healthy leaves. Our ensemble approach achieved an accuracy of 89 % and an F1 score of 0.89 significantly outperforming individual models in both accuracy and consistency across all disease classes. The weights for the ensemble were optimized via systematic grid search validation. We implemented this solution in a mobile application via TensorFlow Lite, thereby allowing real-time detection of the disease with an inference time of approximately 2 seconds. This work shows that ensemble deep learning methods have great potential for agricultural applications and can provide farmers with a user-friendly tool for the early detection and management of diseases in tea plantations.