Hybrid Deep Ensemble of EfficientNetV2 and ConvNeXt for Accurate Paddy Leaf Disease Classification
Year: 2026
Type: conference-paper
Abstract: Aim: The aim of the research is to propose a Hybrid Deep Ensemble Model using soft voting-based score fusion of EfficientNetV2 and ConvNeXt for accurate classification of paddy leaf disease. The proposed model is benchmarked against Inception-ResNetV2 to evaluate improvements in accuracy and ensemble performance. Materials and Methods: Two groups were evaluated. Group 1 employs Inception-ResNetV2, while Group 2 uses the proposed Hybrid Deep Ensemble. A dataset of 3000 paddy leaf images (bacterial leaf blight, brown spot, leaf smut) was pre-processed using normalization, resizing, and augmentation. Both groups trained under identical dataset split (80:20 train/test) and hyperparameter conditions to ensure a fair comparison. Statistical significance was assessed via independent samples t-test (p < 0.05) using SPSS 11.0. Result: Group 1 (Inception-ResNetV2) achieved 92.68% accuracy. Group 2 (Hybrid Ensemble) achieved 98% accuracy, significantly outperforming the baseline (p < 0.05). Conclusion: The Hybrid Deep Ensemble model provides more accurate and stable paddy leaf disease classification than Inception-ResNetV2, demonstrating the advantage of combining complementary CNN architectures via soft voting.
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
SDG: Zero hunger
Open Access status: hybrid
OpenAlex ID: W7204075419