Reviewed by Lexie CornerJul 9 2024
A study published in Plant Phenomics on June 6, 2024, explores the efficacy of Phenogrid, a phenotyping system built for early-stage plant monitoring under biotic stress, with the goal of tackling plant disease resistance.
Distributions of features for batches of resistant and susceptible plants. Image Credit: Plant Phenomics
A study team studied low-cost depth imaging sensors with the goal of automating plant pathology tests. The scientists distinguished resistant and susceptible plants with 97 % accuracy based on cotyledon loss. This approach is 30 times faster than human annotation and is effective in a variety of settings and plant densities.
The unique imaging system, feature extraction approach, and classification model offer a low-cost, high-throughput solution that could be used in decision-support tools and standalone technologies for real-time edge computation.
Selective plant breeding, which began with the domestication of wild plants over 10,000 years ago, has developed to meet the difficulties provided by climate change. Current breeding efforts are aimed at increasing plant resilience to biotic and abiotic challenges, accelerating germination, and boosting nutritional and environmental benefits.
However, developing new varieties, which can take up to ten years, remains a considerable challenge.
In this study, the extraction of spatiotemporal characteristics, such as absolute amplitude (Aabs), relative amplitude (Arel), and drop duration (D), was found to be an efficient strategy for distinguishing between susceptible and resistant plant batches.
The onset (O) characteristic was homogeneous in susceptible plants, whereas resistant plants had a constant three-day onset, which coincided with cotyledon loss. Height signals were less effective, but surface and volume signals revealed significant differences between susceptible and resistant plants.
The statistical tests revealed the importance of most retrieved traits in diagnosing cotyledon loss. This required the deployment of a sophisticated classifier to accomplish efficient batch classification. The Random Forest model got the best classification accuracy (97 %), as well as outstanding performance metrics (MCC: +91 %).
The approach was resilient to inoculation timing fluctuation, retaining performance after up to two hours of desynchronization. Furthermore, simulations showed that lowering the number of plants per batch from 20 to 10 maintained classification performance while doubling throughput.
A visual study demonstrated that direct watering influenced categorization accuracy, implying that automated or subirrigation systems might improve performance even more. The method's success extends to the segregation of additional pathosystems, suggesting strong generalizability and the promise for high-throughput plant pathology diagnosis.
The study’s lead researcher, David Rousseau, stated that the imaging system created, when paired with the feature extraction approach and classification model, provides a comprehensive pipeline with unsurpassed throughput and cost efficiency compared to the state-of-the-art. The system can be used as a decision-support tool, but it is also compatible with a standalone technology that performs computations in real-time at the edge.
Finally, this study demonstrates the effective automation of plant pathology testing utilizing low-cost depth imaging sensors. It discriminates resistance from susceptible plants by cotyledon loss detection with an accuracy of 97 %. The approach is resilient to fluctuations in plant density and desynchronization and hence significantly decreases processing time compared to human annotation.
Future improvements could involve the incorporation of new imaging modalities and the refinement of algorithms for broader applicability, resulting in a speedy, accurate, and cost-effective solution for boosting crop resilience and production.
Journal Reference:
Cordier, M., et. al. (2024) Affordable phenotyping system for automatic detection of hypersensitive reactions. Plant Phenomics. doi:10.34133/plantphenomics.0204