Related Experiment Video
Updated: Jun 12, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
Ontology-driven dual-channel relational graph convolutional network (DCR-GCN) for lettuce leaf phenotype
Ting Li1,2,3, Sheng Wu2,3, Guangjie Qiu2,3
1College of Agricultural Engineering, Shanxi Agricultural University, Taigu, Jinzhong, 030801, China.
Plant Methods
|June 11, 2026
Summary
A new knowledge graph and graph learning framework improves lettuce classification by analyzing complex phenotypic traits. This method enhances germplasm identification and new variety protection for lettuce (Lactuca sativa L.).
Area of Science:
- Agricultural Science
- Computer Science
- Genetics
Background:
- Lettuce (Lactuca sativa L.) exhibits significant diversity in leaf traits, complicating germplasm identification and variety protection.
- Existing phenotypic analysis methods struggle to represent and utilize semantic relationships between traits.
- Accurate phenotypic analysis is crucial for lettuce breeding and intellectual property protection.
Purpose of the Study:
- To develop a novel knowledge graph-enhanced graph learning framework for analyzing lettuce phenotypic traits.
- To improve the accuracy and interpretability of lettuce type classification.
- To provide methodological support for precise germplasm identification and new variety protection.
Main Methods:
- Phenotypic traits were extracted from lettuce leaf images.
- A Lettuce Leaf Phenotypic Trait Knowledge Graph (LLPT-KG) was constructed using International Union for the Protection of New Varieties of Plants (UPOV) standards.
- A Dual-Channel Relational Graph Convolutional Network (DCR-GCN) was employed to integrate trait features and graph structures for classification.
- Node- and edge-level importance analyses were conducted for interpretability.
Main Results:
- The DCR-GCN framework achieved high accuracy (0.94) and Macro-F1 score (0.94) in lettuce type classification.
- The proposed method significantly outperformed the Relational Graph Convolutional Network (R-GCN) baseline, improving accuracy by ~9% and Macro-F1 by ~10%.
- Key phenotypic traits and semantic relationships relevant to type discrimination were identified.
Conclusions:
- Integrating knowledge graphs with graph neural networks effectively captures complex phenotypic relationships in lettuce.
- The developed framework offers a robust solution for precise lettuce germplasm identification and digital phenotypic evaluation.
- This approach supports digital-assisted pre-screening for Distinctness, Uniformity, and Stability (DUS) testing and new variety protection.
