Achieving Near-Perfect Performance: A Neuro-Fuzzy-Enhanced Meta-Learning Model for Automated Severity-Graded Plant Disease Diagnosis

Abstract

Plant diseases threaten global food security and agricultural productivity. While deep learning offers diagnostic potential, real-world deployment is limited by data scarcity, algorithmic bias, and lack of interpretability. This paper overcomes these barriers with the development of a novel Neuro-Fuzzy Enhanced meta-learning model integrating a SqueezeNet embedder for efficient feature extraction and an adaptive neuro-fuzzy inference system for interpretable, severity-graded prediction. Trained and validated on 51,000 images of diseased cassava, maize, and banana, the model achieves near perfect exceptional performance of 99.92% accuracy, 99.8% Precision, 99.7% Recall, and 98.9% F1-score. It maintains a 92.0% F1-score in 5-shot learning, demonstrating rapid adaptability to novel diseases. The model provides reliable severity assessment with a minimum severity score of 2.4 (scale 1.0-4.0) and operates efficiently on edge devices. This findings contained in this paper delivers a scalable, accurate, and interpretable AI solution, highlighting the critical roles of optimal feature selection, meta-learning, and neuro-fuzzy logic in advancing practical, trustworthy agricultural diagnostics for global food security.