Abstract:
This study presents a Systematic Literature Review (SLR) on the early detection and diagnosis of
coconut leaf diseases in Sri Lanka’s Coconut Triangle, a region critical to the nation’s agricultural
economy. The review systematically evaluates published research on deep learning-based models,
image classification techniques, and hybrid approaches that integrate environmental data for more
reliable diagnosis. To ensure methodological rigor, the study followed the PRISMA (Preferred
Reporting Items for Systematic Reviews and Meta-Analyses) framework. Relevant studies were
retrieved from scholarly databases including IEEE Xplore, ScienceDirect, Springer, Scopus, and
Google Scholar using predefined search strings. Strict inclusion and exclusion criteria were applied
to filter peer-reviewed articles published between 2016 and 2025, ensuring relevance and
reproducibility. A total of 20 papers met the criteria for final review, covering a diverse range of
approaches such as Convolutional Neural Networks (CNNs), transfer learning models (VGG16,
ResNet, EfficientNet), ensemble methods (CNN–SVM), and IoT-assisted systems. The synthesis
of findings highlights several insights: transfer learning-based CNNs achieved over 90% accuracy,
even for subtle early-stage symptoms. Hybrid approaches combining CNNs with classifiers like
SVMs improved precision, reducing false positives and enhancing robustness across coconut
varieties. Integrating environmental data (humidity, rainfall, temperature) increased accuracy by 6–
10%, stressing the need for region-specific models in Sri Lanka’s diverse agro-climatic zones.
However, most systems lack large-scale field validation, mobile deployment, and consideration of
smallholder farmer resource constraints. Key research gaps include limited Sri Lanka-specific
datasets, inadequate focus on multi-disease co-infections, and the absence of farmer-oriented
mobile and IoT tools for real-time disease monitoring. Finally, the study highlights future directions
such as building larger and varietal-specific datasets, experimenting with advanced architecture
(EfficientNet, Vision Transformers), and developing lightweight, offline-capable mobile
applications to ensure accessibility for rural farmers. The integration of IoT sensors and climate
forecasting models is also recommended to move from disease detection towards predictive and
preventive farming strategies. This review contributes to the body of knowledge by not only
consolidating existing research but also providing actionable insights for researchers, policymakers,
and agricultural technology developers to ensure sustainable disease management and resilience in
Sri Lanka’s coconut sector.
