Preskoči na sadržaj
Moj profil

izvorni znanstveni rad · 2026

An Explainable Neuro-Fuzzy Multimodal Fusion Approach for Lightweight Edge-Based Plant Disease Detection in Precision Agriculture

Domagoj Tuličić, pred.

Sažetak

Plant diseases pose a significant threat to global food security, causing annual crop losses estimated at 20–40% and demanding intelligent monitoring systems for early detection and intervention. Automated disease classification using deep learning has shown remarkable success on benchmark datasets; however, existing approaches predominantly rely on single-modality RGB imagery and lack the interpretability essential for practical agronomic decision-makingqueryThis is to inform you that corresponding author has been identified as per the information available in the Copyright form.. Current fusion strategies fail to adequately model uncertainty and inter-modal dependencies inherent in real-world agricultural environments. This research proposes a novel neuro-fuzzy multimodal fusion architecture that integrates lightweight depthwise separable convolutional networks with adaptive neuro-fuzzy inference systems for edge-deployable plant disease classification. The framework processes RGB, thermal, and multispectral modalities through parallel feature extractors, fusing representations via learnable Gaussian membership functions optimized using Particle Swarm Optimization. Experimental evaluation on the PlantVillage dataset comprising 20,638 images across 15 disease classes demonstrates 99.13% test accuracy with only 950,968 parameters, outperforming comparable lightweight architectures while maintaining a compact 3.63 MB footprint. The proposed approach bridges the gap between black-box deep learning and explainable agricultural decision support, offering a practically deployable solution for resource-constrained edge devices in precision farming applications.

Ključne riječi

neuro-fuzzy systems multimodal fusion plant disease detection edge computing lightweight convolutional neural networks