Wheat Sunn Pest Detection and Species Classification Using Deep Ensemble Learning

This project presents SunnPest, an advanced deep learning system for automated detection of Sunn Pest damage and classification of wheat varieties. Addressing critical agricultural challenges in wheat crop protection, the solution replaces traditional multi-stage pipelines with three high-performance supervised models: ConvNeXt-Base, EfficientNetV2-M, and their soft-voting Ensemble, achieving ≥90% accuracy. The models leverage pretrained backbones fine-tuned on a dataset of 170+ wheat images across 6 species (bezostaja, ekiz, mufitbey, nacibey, sonmez-2001, tosunbey) and two conditions (healthy/damaged). A robust pipeline incorporating stratified 70/15/15 splitting, aggressive data augmentation, and Test-Time Augmentation (TTA) ensures strong generalization, delivering perfect damage detection (100% accuracy, AUC 1.0) and competitive species classification (~85% ensemble accuracy) on the test set. A user-friendly Flask-based web application enables real-time inference with image upload, ensemble predictions, annotated visualization, and per-model breakdown. Features include secure user authentication via MySQL, TTA toggling, and performance dashboards.

Keywords: Wheat Sunn Pest Detection, ConvNeXt, EfficientNetV2, Ensemble Learning, Test-Time Augmentation (TTA), Deep Learning, Image Classification, Agricultural AI, Precision Agriculture, Flask Web Application.