LeTra: Tracking Plant Growth with AI and Smart Overlap Measures

Jim Crocker
18th January, 2024

LeTra: Tracking Plant Growth with AI and Smart Overlap Measures

Graphical abstract of the approach used from the study, consisting of two main stages.

Image adapted from: Jurado-Ruiz et al. / CC BY (Source)
Improving crop yields relies heavily on understanding how plants convert light into energy through photosynthesis. Traditionally, assessing photosynthetic efficiency has been a slow, manual process, limiting the speed of plant breeding programs. Recent advancements in high-throughput phenotyping (HTP) – technologies that allow for rapid, detailed measurement of plant traits – have begun to address this bottleneck[2]. However, while HTP systems can efficiently measure photosynthesis at the whole-plant level, detailed analysis at the individual leaf level has remained challenging, often requiring significant manual effort. Researchers at CRAG (Center for Research in Agricultural Genomics) have developed a new automated method to overcome this limitation[1]. Their work focuses on accurately identifying and tracking individual leaves over time using computer vision, specifically a technique called Mask R-CNN, a type of convolutional neural network. This network is ‘trained’ to recognize and outline leaves in images taken from directly above the plants. The key innovation lies in adapting this technology to automatically follow the growth and changes of each leaf throughout its lifespan. The challenge with tracking leaves isn’t simply identifying them in a single image; it’s maintaining that identification as the plant grows and leaves move or become partially obscured. To address this, the researchers used a metric called ‘intersection over union’ (IoU) to determine how well the predicted leaf outline in one image matches the same leaf in a subsequent image. This, combined with a metric called Higher Order Tracking Accuracy (HOTA), allowed for robust tracking even with overlapping leaves or changes in leaf shape. The method was tested on images of Arabidopsis thaliana, a small flowering plant commonly used in plant research, with a dataset of 523 leaves at different developmental stages. The results were highly accurate, achieving a mean F-score of 0.956 for detecting leaves and 0.844 for accurately outlining their shape. Tracking 191 leaves across nine plants yielded 84.29% correct tracking, with a HOTA score of 0.846. Importantly, the researchers found that even with a relatively small amount of training data, the system performed well, though expanding the dataset further improved accuracy. This level of detail allows researchers to investigate how photosynthesis varies across different leaves on the same plant. Their case study revealed that leaf age and position on the plant (leaf order) both influence photosynthetic capacity and how leaves respond to different light levels. This aligns with previous research demonstrating that photosynthetic efficiency can vary significantly depending on environmental factors[3][4]. For example, studies have shown that different barley varieties exhibit varying tolerances to high and low light conditions, impacting their photosynthetic performance[4]. Furthermore, the study demonstrated that leaf-dependent photosynthesis isn’t solely determined by environmental factors but also by the plant’s genetic makeup. This is crucial for breeding programs aiming to improve crop yields, as it highlights the importance of considering how genetic variation influences photosynthetic efficiency at the individual leaf level. The ability to rapidly and accurately assess these variations, as enabled by the new method, represents a significant step forward in plant phenotyping[5]. The researchers have made their datasets and code publicly available, encouraging further development and application of this technology within the plant science community. This open-source approach is intended to accelerate the adoption of advanced computer vision techniques in phenotyping platforms, ultimately contributing to more efficient crop improvement strategies.

AgricultureBiotechBiochem

References

Main Study

1) LeTra: a leaf tracking workflow based on convolutional neural networks and intersection over union.

Published 17th January, 2024

https://doi.org/10.1186/s13007-024-01138-x

Related Studies

2) High-Throughput Phenotyping: A Platform to Accelerate Crop Improvement.

https://doi.org/10.1007/s43657-020-00007-6

3) Chlorophyll Fluorescence Imaging as a Tool for Evaluating Disease Resistance of Common Bean Lines in the Western Amazon Region of Colombia.

https://doi.org/10.3390/plants11101371

4) Fluorescence parameters as early indicators of light stress in barley.

https://doi.org/10.1016/j.jphotobiol.2012.03.009

5) Phenomics for photosynthesis, growth and reflectance in Arabidopsis thaliana reveals circadian and long-term fluctuations in heritability.

https://doi.org/10.1186/s13007-016-0113-y


Related Articles

An unhandled error has occurred. Reload 🗙