QPlantNet: A Quantum-Inspired Diffusion-Augmented Capsule Network for Accurate Multi-Crop Plant Disease Classification

Abstract

Plant diseases significantly reduce agricultural productivity, threaten global food security, and cause substantial economic losses worldwide. Early and accurate identification of crop diseases is therefore essential for implementing timely disease management strategies and improving agricultural sustainability. However, conventional deep learning-based plant disease classification approaches often suffer from limited training data, high intra-class similarity, complex inter-class variations, poor generalization across diverse crop species, and limited interpretability. To overcome these challenges, this paper proposes QPlantNet, a Quantum-Inspired Diffusion-Augmented Capsule Network for accurate multi-crop plant disease classification. The proposed framework integrates a diffusion-based generative augmentation module, a capsule network-based feature extraction mechanism, and a quantum-inspired optimization strategy into a unified architecture. The diffusion augmentation module generates realistic synthetic disease images by learning complex visual distributions of infected crop regions, thereby improving dataset diversity and reducing overfitting. The capsule network effectively preserves hierarchical and spatial relationships among disease symptoms, enabling robust recognition of visually similar disease patterns. Furthermore, the quantum-inspired optimizer enhances parameter exploration, accelerates convergence, and improves generalization by avoiding local optima during model training. To improve model transparency, Gradient-weighted Class Activation Mapping (Grad-CAM) is incorporated for visual explanation of disease prediction decisions. Extensive experiments on benchmark multi-crop plant disease datasets demonstrate that QPlantNet consistently outperforms conventional CNN, ResNet, EfficientNet, Vision Transformer, and Capsule Network models. The proposed framework achieves an overall classification accuracy of 99.10%, 98.90% precision, 99.00% recall, and an F1-score of 98.95%. Additional evaluation using ROC curves, Precision–Recall analysis, and explainability assessment confirms the robustness, reliability, scalability, and practical applicability of the proposed framework for intelligent plant disease diagnosis and precision agriculture applications.

Plant phenotyping relevance

植物病害画像から病徴を分類する新規画像解析手法を開発・評価しており、植物の病害状態を観測する方法が研究の中心です。

Code and data availability

The article describes the QPlantNet framework and benchmark results but contains no data availability statement, no public dataset identifiers, no repository links, and no code or model release. No paper-specific public assets are present in the supplied blocks.

No evidence-backed public reproduction asset is currently recorded.