Advanced Hybrid Deep Learning Framework for Multi-Scale Apple Disease Detection and Classification with Ensemble Learning
Procedia Computer Science · 2026 · 283 · 5378-5388
DOI: 10.1016/j.procs.2026.06.588 ↗Kurz gefasst
Fachartikel in der Zeitschrift Procedia Computer Science, 2026. Verfasst von Akey Sungheetha und 1 weiteren Autor:innen. Die Arbeit nennt 90 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
This paper proposes a multi-modal ensemble framework for apple leaf disease Classification that integrates a bottleneck ResNet-101 backbone augmented with Squeeze-and-Excitation (SE) attention, Feature Pyramid Network (FPN), and four-level Spatial Pyramid Pooling (SPP) to expand feature representations from 2,048-d to 102,400-d (2048×50). Three-modality fusion (RGB, HSV, Augmented) with temperature-scaled calibration (T=2.0, w RGB =0.72, w HSV =0.68, w AUG =0.64) is applied across 2,848 apple leaf images spanning six disease categories, augmented to 7,292 training samples. The proposed ensemble achieves 97.77% accuracy, macro-F1 of 0.953, and macro-average AUC of 0.981 on a 539-sample test set at 52 ms per image, surpassing EfficientNet-B4 (93.0%) by 4.77 percentage points. Ablation confirms cumulative gains from SE attention (+1.6%), FPN (+1.3%), SPP (+0.8%), and ensemble calibration (+0.87%) over the 92.7% baseline, validating each component’s indispensable contribution to precision agriculture deployment. Experimental validation on Plant Village Dataset containing 3642 images across six disease categories demonstrates superior performance with precision of 96.4 percent, recall of 97.2 percent, and F1-score of 96.8 percent. The proposed framework processes images at 42 frames per second enabling real-time deployment in precision agriculture applications. Comparative analysis reveals 8.3 percent improvement over standalone deep learning models and 12.7 percent enhancement compared to traditional machine learning approaches
Zitieren
APA
Sungheetha, A., & R, R. S. (2026). Advanced Hybrid Deep Learning Framework for Multi-Scale Apple Disease Detection and Classification with Ensemble Learning. Procedia Computer Science, 283, 5378-5388. https://doi.org/10.1016/j.procs.2026.06.588
Vancouver
Sungheetha A, R RS. Advanced Hybrid Deep Learning Framework for Multi-Scale Apple Disease Detection and Classification with Ensemble Learning. Procedia Computer Science. 2026;283:5378-5388. doi:10.1016/j.procs.2026.06.588
BibTeX
@article{Sungheetha2026,
author = {Sungheetha, Akey and R, Rajesh Sharma},
title = {Advanced Hybrid Deep Learning Framework for Multi-Scale Apple Disease Detection and Classification with Ensemble Learning},
journal = {Procedia Computer Science},
year = {2026},
volume = {283},
pages = {5378--5388},
doi = {10.1016/j.procs.2026.06.588}
}
EUVIMED