Abstract:
The rubber industry is a vital component of the Sri Lankan economy, supporting thousands of
livelihoods. However, leaf diseases can reduce rubber yield by up to 45%, posing a significant
threat to productivity and income. Traditional methods of disease detection are time-consuming
and inaccurate, leading to delayed interventions and increased crop losses. To address this issue,
this research investigates the use of deep learning for automated rubber leaf disease diagnosis,
enabling quicker and more accurate identification to minimize losses. The study evaluates
multiple Convolutional Neural Network (CNN) models including YOLOv8, VGG16 and ResNet50,
using a dataset of rubber leaf images collected from Sri Lankan plantations. Data augmentation
techniques were applied to enhance dataset diversity, while preprocessing steps were used to
optimize model performance. Transfer learning was used to improve efficiency and reduce the
need for extensive training data. Among the models tested, VGG16 demonstrated the most
balance of accuracy (93.65%), F1-score (0.9333) and computational efficiency, making it the
optimal choice for real-world deployment. YOLOv8 also performed competitively, offering fast
inference times (7.65s) with a lightweight architecture (1.44M parameters). These findings
highlight the potential of AI-driven disease detection to improve farm management, reduce crop
losses, and enhance agricultural sustainability. This research contributes to climate-smart
agriculture by providing an innovative, data driven solution for early disease detection, helping
rubber farmers adapt to climate-induced disease pressures. By incorporating deep learning into
agriculture, this study supports resilient and sustainable rubber cultivation in Sri Lanka and other
comparable agro-climatic zones.
