Leaf area index estimation in maize breeding trials from RGB imagery and machine learning algorithms

In this study, a novel AI-based model, based on a convolutional neural network, has been developed and validated for leaf area index (LAI) estimations of maize using downward facing RGB images taken from a high-throughput plant phenotyping platform. Ground truth LAI values were obtained by measuring the leaf dimensions of selected plants and using an allometric relationship. The results obtained from the model approach were compared against ground truth values and LAI estimations from a classic indirect method based on hemispherical images and gap fraction theory. The model showed good performance with ground truth values and LAI estimations by the classic indirect method. This represents a major advance in LAI estimates in maize plots when compared to previous methods, particularly in terms of processing time and equipment cost.