Coconut palm diseases pose a significant threat to tropical economies and the livelihoods of smallholder farmers. While centralized deep learning models have achieved high diagnostic accuracy, their deployment in real-world agricultural settings is hindered by logistical challenges and data privacy concerns. This paper proposes the CocoSyn framework, which leverages the high-performance DeepSeqCoco model to establish a robust diagnostic baseline for stable coconut disease diagnosis using Federated Learning (FL). To ensure true deployment readiness, a factor often overlooked in agricultural Artificial Intelligence (AI), we introduce the Stability Assurance for Federated Edge (SAFE) Protocol. This rigorous multi-seed evaluation standard verifies training stability and exposes outlier risks. Applying the SAFE protocol, we conduct a critical comparative analysis of the Adaptive Moment Estimation (Adam) and Stochastic Gradient Descent (SGD) optimizers. Our findings reveal that while both can achieve high accuracy, SGD is prone to catastrophic failure, a risk only detectable through our stability-focused protocol. The CocoSyn framework is successfully validated by the SAFE protocol, demonstrating its superior algorithmic stability and resilience with the Adam optimizer. This work establishes a new, more rigorous standard for vetting FL systems, ensuring they are not only accurate but demonstrably reliable for real-world deployment.
Bommineni et al. (Thu,) studied this question.