Towards High Throughput Stress Phenotyping in Soybeans Using Machine Learning.

Towards high throughput stress phenotyping in soybeans using machine learning

Presenter: Arti Singh

Plant phenotyping for accurate and precise trait collection is becoming more complex and sophisticated due to the large influx of different ground and aerial platforms, sensors, and time series data collection on a large set of genotypes. High-throughput phenotyping (HTP) has unlocked new prospects for non-destructive field and controlled conditions based phenotyping for a large number of traits, including biotic and abiotic stress traits, which are critical for genetic enhancement. Machine-learning (ML) tools have been shown to work with data obtained from complex and integrated phenotyping platforms that solve the identification, classification, quantification, and prediction paradigm to generate insights that were previously not possible.

In this presentation, two examples will demonstrate the application of ML for high throughput stress phenotyping. In the first example, advances in imaging technology, data analytics, and ML are leveraged to enable automated and fast phenotyping and subsequent decision support in soybean for iron deficiency chlorosis. The phenotyping and information output work flow includes imaging → data storage → data curation → machine learning enabled classification of genotypes → development of decision support tools which include smartphone apps. In the second example, an automated framework to identify and classify biotic and abiotic stresses on soybean using Deep Convolutional Neural Networks (DCNN) is presented. Results show that the trained model is able to efficiently differentiate between eight different soybean stresses and reasonably classify each stress into categories for a qualitative rating, used predominantly in scouting and breeding applications. The trained model is amenable to easy export to a mobile imaging platform, utilizing the visual sensors in a dedicated low-cost imaging device or smartphone camera, and can perform an efficient and robust online detection of soybean stresses in real-time.