Vision-driven retrieval-augmented decision support for plant disease diagnosis using EfficientNetV2 and large language models

Accurate and timely plant disease diagnosis is essential for sustainable agricultural production and precision farming. However, most existing agricultural decision support systems treat visual disease diagnosis and agronomic recommendation generation as separate processes, either providing image classification results alone or generating recommendations with large language models that are not grounded in domain-specific knowledge. To address this limitation, this study introduces EffiToRag, an end-to-end decision support framework that integrates an EfficientNetV2-based visual diagnosis module with a Retrieval-Augmented Generation (RAG) architecture. Rather than relying solely on free-text queries for disease-level retrieval, the proposed approach uses the disease class predicted by EfficientNetV2 as the primary key for retrieving the corresponding record from a structured agronomic knowledge base constructed from peer-reviewed scientific publications and official phytopathological guidelines. The visual diagnosis module was trained using Progressive Training and Class-Aware Tiered Augmentation (CATA) to account for differences in class representation. On a held-out test set comprising 11,621 images across 60 plant health classes, EfficientNetV2-M achieved 99.13% accuracy and a Macro F1 score of 0.9889, with an inference time of 4.39 ms per image on the evaluated GPU. For 500 user queries, the system selected query-relevant knowledge fields with a precision of 0.972 and a recall of 0.985. The factual-consistency assessment combined GPT-4-based evaluation and manual expert review of a random sample of 100 diagnostic reports and yielded a mean factual-consistency score of 0.99. The experts also assigned these reports a mean applicability score of 4.85/5.0. These findings show that visual disease classification, structured agronomic knowledge retrieval, and LLM-based recommendation generation can be integrated within a single decision support pipeline for precision agriculture.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-18
DOI
https://doi.org/10.1038/s41598-026-78385-6
Primary Topic
Smart Agriculture and AI
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Vision-driven retrieval-augmented decision support for plant disease diagnosis using EfficientNetV2 and large language models

Serpil Aslan, Vedat Tümen, Erdal Akin, Fatih Kaya
3 links
Scientific Reports
Smart Agriculture and AI
article

Vision-driven retrieval-augmented decision support for plant disease diagnosis using EfficientNetV2 and large language models

Serpil Aslan, Vedat Tümen, Erdal Akin, Fatih Kaya
article en

Abstract

Accurate and timely plant disease diagnosis is essential for sustainable agricultural production and precision farming. However, most existing agricultural decision support systems treat visual disease diagnosis and agronomic recommendation generation as separate processes, either providing image classification results alone or generating recommendations with large language models that are not grounded in domain-specific knowledge. To address this limitation, this study introduces EffiToRag, an end-to-end decision support framework that integrates an EfficientNetV2-based visual diagnosis module with a Retrieval-Augmented Generation (RAG) architecture. Rather than relying solely on free-text queries for disease-level retrieval, the proposed approach uses the disease class predicted by EfficientNetV2 as the primary key for retrieving the corresponding record from a structured agronomic knowledge base constructed from peer-reviewed scientific publications and official phytopathological guidelines. The visual diagnosis module was trained using Progressive Training and Class-Aware Tiered Augmentation (CATA) to account for differences in class representation. On a held-out test set comprising 11,621 images across 60 plant health classes, EfficientNetV2-M achieved 99.13% accuracy and a Macro F1 score of 0.9889, with an inference time of 4.39 ms per image on the evaluated GPU. For 500 user queries, the system selected query-relevant knowledge fields with a precision of 0.972 and a recall of 0.985. The factual-consistency assessment combined GPT-4-based evaluation and manual expert review of a random sample of 100 diagnostic reports and yielded a mean factual-consistency score of 0.99. The experts also assigned these reports a mean applicability score of 4.85/5.0. These findings show that visual disease classification, structured agronomic knowledge retrieval, and LLM-based recommendation generation can be integrated within a single decision support pipeline for precision agriculture.

Scientific Reports
Malmö University (SE), Bitlis Eren University (TR), Malatya Turgut Özal Üniversitesi (TR), Turgut Özal University (TR)
Malmö Högskola
Zero hunger
Smart Agriculture and AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon