Answers are grounded in this paper and cite their sources.
Coconut diseases significantly affect agricultural productivity, while manual disease identification is time-consuming and error-prone. This study proposes a dual-optimizer-driven hybrid ensemble for automated coconut disease recognition. Deep features were extracted and fused from three pretrained CNN architectures: InceptionV3, EfficientNetB0, and MobileNetV2. The Firefly Algorithm (FA) and Archerfish Hunting Optimizer (AHO) were sequentially employed to optimize feature fusion and classifier configurations. A leakage-controlled evaluation protocol using exact hashing, perceptual hashing, and deep-feature clustering was implemented to ensure plant-disjoint data partitioning. Data augmentation and SMOTE were restricted to the training partition. Experiments on 5,798 field images covering five coconut disease categories achieved a mean test accuracy of 97.13% (95% CI: 96.56%–97.70%) and a macro-F1 score of 96.32% across five repeated runs. Nested group-aware cross-validation achieved 97.99% accuracy. The leakage audit revealed an approximately 1.6 percentage-point overestimation when same-plant images were not grouped. The FA–AHO search performed comparably to random search under an identical evaluation budget. The deployed model achieved 21.8 ms inference latency on an NVIDIA Tesla T4 GPU. The proposed framework provides a leakage-controlled and interpretable approach to coconut disease recognition. Although optimizer and feature-fusion gains were not statistically significant on this near-saturated benchmark, the study establishes a reproducible evaluation framework and demonstrates practical computational efficiency. Independent multi-site validation remains necessary before field deployment.
No takes yet. Share an insight, caveat, or question.
Arulgnanaselvi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: