Large Adaptive Fusion Ensemble
Abstract
Accurate and timely identification of plant diseases is critical for ensuring global food security and promoting sustainable agriculture. This document presents PlantNet, a state-of-the-art, production-ready deep learning system for plant disease classification. PlantNet utilizes a sophisticated ensemble architecture that leverages the complementary strengths of four powerful computer vision models: ResNet152, EfficientNetB4, Vision Transformer (ViT-B/16), and Swin Transformer. By fusing the outputs of these parallel backbones through an adaptive weighted averaging mechanism, the model achieves superior robustness and generalization. Trained on the extensive PlantVillage dataset, PlantNet attains an accuracy of 97.0% across 38 distinct disease classes. Optimized for high-throughput inference (45 images/second) and available in multiple deployment formats (PyTorch, ONNX, TorchScript), PlantNet is designed for seamless integration into real-world agricultural workflows. This work demonstrates the power of ensemble methods in creating highly reliable diagnostic tools to support farmers and advance AI-driven agricultural practices.