A Hybrid Deep Learning Model for Improving Tea Leaf Disease Detection: Overcoming Challenges in Sri Lanka’s Tea Industry.

Abstract

In the Sri Lankan tea industry, tea leaf diseases affect yield and quality. Therefore, this study proposes a hybrid deep learning model that combines MobileNet, VGG16, and ResNet50 to improve the accuracy, reliability, and robustness in detecting tea leaf disease. Each model is unique in that MobileNet contributes with its lightweight and computationally efficient nature, VGG16 extracts deep features hierarchically, and ResNet50 relies on residual connections to improve learning. However, each of them suffers from common problems in capturing complex features effectively, a higher computational load, and a longer training time while working with large or diverse datasets. Thus, the hybrid model will be able to achieve maximum performance in classification by overcoming the limitations of each model through strategic combinations of the architectures. 5,000 images of tea leaves were collected from Kaggle, labeled, and preprocessed by resizing, normalizing, and augmenting to improve the generalization and avoid overfitting. The experimental performance of individual models, as shown by the results, revealed that MobileNet classified tea leaf diseases with an accuracy of 83.9%, ResNet50 with 96.30%, and VGG16 with 93.63%. These models are to be integrated into an organized hybrid framework through ensemble learning methods, which are intended to provide optimized robustness, reduce prediction errors, and improve model interpretability for applications in agriculture. This study fills an important gap in the literature by investigating how the hybrid approach outperforms both individual models and demonstrates that the synergistic use of multiple architectures enhances disease detection accuracy and consistency. Such an approach may revolutionize large-scale agricultural disease monitoring by reducing dependency on manual inspections, which are often prone to human error. In the future, fine-tuning of the hybrid model will be done through the implementation of fine-tuning strategies, optimization of hyperparameters, and testing on real-world deployments to enhance its feasibility for tea plantations.