Exemplar-based data generation and leaf-level analysis for phenotyping drought-stressed poplar saplings

A research team has developed a novel method combining computer vision and deep learning to phenotype drought-stressed poplar saplings, achieving 99% accuracy in variety identification and 76% accuracy in stress-level classification. This high-accuracy phenotyping approach leverages instance segmentation and multitask learning, offering precise drought-stress detection. The methods proposed hold significant potential for drought-resistant poplar screening and precise irrigation decision-making, fostering advancements in agricultural technology and plant stress management.

Poplar (Populus L.) is a fast-growing forest tree valued for its wood and role in protective forests. Current research focuses on enhancing woody biomass production despite abiotic and biotic stresses, with drought stress being a significant factor that impedes growth by affecting material transport and photosynthesis. However, traditional methods for identifying water-deficient plants or selecting drought-resistant varieties are inefficient and inaccurate.

A study (DOI: 10.34133/plantphenomics.0205) published in Plant Phenomics on 21 Jun 2024, explores innovative computer vision and deep learning technologies to improve drought-stress detection and phenotyping in poplar saplings.

The research utilized instance segmentation and leaf posture digitalization to analyze poplar saplings. The FasterRCNN model was used to segment leaves, midveins, and petioles, outperforming YOLO models in some aspects. The segmentation accuracy was evaluated using AP0.5 values, with FasterRCNN showing higher performance for leaf segmentation and YOLO excelling in midvein and petiole detection. The angles α and β, indicating leaf growth posture, were calculated, revealing errors only in incomplete segmentations. The mean absolute errors for petiole and midvein angle calculations were 10.7° and 8.2°, respectively. These results were validated with a new dataset, showing most errors within a range of [-5°, +5°]. The study confirmed the effectiveness of using segmentation models trained on a simulated dataset for accurate leaf posture analysis, despite some deviations under severe drought conditions. Further analysis indicated that the midvein's horizontal inclination angle was more affected by drought stress than the petiole's angle, proving the leaf posture calculation method's value in plant status analysis.

According to the study's lead researcher, Huichun Zhang, “The plant phenotyping methods presented in this study could be further used for drought-stress-resistant poplar plant screening and precise irrigation decision-making.”

In summary, this study used computer vision and deep learning to phenotype drought-stressed poplar saplings, focusing on leaf posture calculation and stress-level identification. These methods significantly reduced manual annotation costs and demonstrated the potential for precise drought stress detection. Future research will focus on improving segmentation accuracy and expanding these techniques to other plant species for enhanced agricultural management.

###

References

DOI

10.34133/plantphenomics.0205

Original Source URL

https://doi.org/10.34133/plantphenomics.0205

Funding information

This work is supported by the National Key Research and Development Program of China (2023YFE0123600), the National Natural Science Foundation of China (NSFC32171790, 32171818, and 62305166) and the Jiangsu Province Agricultural Science and Technology Independent Innovation Fund Project (CX(23)3126).

About Plant Phenomics

Plant Phenomics is an Open Access journal published in affiliation with the State Key Laboratory of Crop Genetics & Germplasm Enhancement, Nanjing Agricultural University (NAU) and published by the American Association for the Advancement of Science (AAAS). Like all partners participating in the Science Partner Journal program, Plant Phenomics is editorially independent from the Science family of journals. Editorial decisions and scientific activities pursued by the journal's Editorial Board are made independently, based on scientific merit and adhering to the highest standards for accurate and ethical promotion of science. These decisions and activities are in no way influenced by the financial support of NAU, NAU administration, or any other institutions and sponsors. The Editorial Board is solely responsible for all content published in the journal. To learn more about the Science Partner Journal program, visit the SPJ program homepage.