Explainable AI -driven deep learning model for the root image-based classification of the medicinal plants Saposhnikovia divaricata, Glehnia littoralis, and Peucedanum japonicum

Read full text Summarise with VIRAT
Authors: Hyein Lee, Yu-Jin Jeon, So Hyeon Park, Ho-Youn Kim, Da Hye Ryu, Jung-Ok Kim, Joo-Young Kim, Dae-Hyun Jung
Cited by: 0
References: 35

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Understanding how ecological functional traits are encoded and recognized by data-driven models remains a critical challenge in ecological informatics. In particular, root anatomical traits such as xylem coloration and phloem texture reflect plant adaptation to environmental conditions, yet their quantitative interpretation from image data is still limited. In this study, we propose a quantitative explainable artificial intelligence (XAI)-based framework to investigate whether deep learning models capture ecologically relevant root phenotypic traits in root cross-sectional images of three morphologically similar but ecologically distinct medicinal plants ( Saposhnikovia divaricata , Glehnia littoralis , and Peucedanum japonicum ). Rather than focusing on model performance alone, we design a structured analytical pipeline that links model attention to trait-relevant anatomical structures and evaluates their consistency with ecologically interpretable trait signals. Using region-based masking and activation analysis, we quantify how models attend to xylem and phloem tissues and examine how controlled manipulation of color and texture features influences trait-specific attention responses. The results show that model attention is systematically associated with anatomical regions linked to plant functional traits, and that architectures designed for fine-grained recognition exhibit stronger alignment with ecologically meaningful features. These findings demonstrate that interpretable deep learning can be extended beyond prediction to serve as a framework for extracting and validating ecological signals embedded in plant functional traits. The proposed approach provides a pathway for integrating image-based AI with ecological trait analysis, enabling quantitative linkage between image-derived phenotypic traits and environmental adaptation in ecological informatics.

Fields of Study & Subjects

Explainable Artificial Intelligence (XAI) Smart Agriculture and AI Advanced Neural Network Applications

Publication details

Journal / source
Ecological Informatics
Volume
95
Pages
103743–103743
Publication date
31 March 2026
Language
EN