Applied Artificial Intelligence in High Throughput Plant Phenotyping for Recognition of Arabidopsis Ecotypes

https://ipmb-2023.p.asnevents.com.au/days/2024-06-25/abstract/103921  

Image-based high-throughput plant phenotyping utilises innovative imaging infrastructures to conduct automated and non-invasive time-series measurements of morphological, physiological, biochemical, and ecological traits of plant species. The primary aim is to determine the structure, performance, and tolerance to limitations of an individual plant or group of plants in a laboratory, glasshouse, or field environment. Recognition of plant ecotypes plays an important role while evaluating, selecting, and producing cultivars. However, it is challenging, time-consuming and nearly impossible to correctly identify multiple ecotypes purely based on visual inspection or through manual measurements of phenotypic traits such as plant area. This research aims to classify various ecotypes of the experimental plant known as Arabidopsis thaliana grown under controlled conditions in indoor high-throughput phenotyping environments. 40 ecotypes are scanned from the top-view on a daily basis taking four images using high-resolution RGB and thermal camera. Upon data collection, all images are organised and stored in the database for subsequent use. After that, several deep learning models are optimised, trained, and evaluated to conduct the classification of various ecotypes considering a single or sequence of images at the input of the model. For the sequence of images, models are trained and evaluated in order to desirable performance for specific days after sowing. Owing to the high variability among replicates within ecotypes, certain samples have to be excluded by extracting general leaf phenotypic traits and considering morphological similarity. Finally, applied deep learning models demonstrated a high level of accuracy and precision during the ecotype recognition task, which provides an opportunity to utilise trained models in different indoor environments, for instance, a glasshouse.