An RGB deep-neural network approach for high-throughput phenotyping of Fusarium head blight in wheat

Fusarium head blight (FHB) in wheat is an economically important disease which can cause yield losses exceeding 50%. Breeding for host resistance remains the most effective disease control method; however, time, labor, and human subjectivity during disease scoring limits selection advancements. In this study we describe an innovative, high-throughput phenotyping rover for capturing in-field RGB images and a deep neural network pipeline for wheat spike detection and FHB disease quantification. The image analysis pipeline detects wheat spikes from images collected by the phenotyping rover under variable field conditions, segments those spikes and the amount of diseased tissue in the spikes, and quantifies disease severity as the region of intersection between the spike and disease masks. To validate disease inferences, individual spike and plot aggregate FHB estimates from the pipeline were compared with visual disease scores from the field and on images. The precision and throughput of the pipeline surpassed traditional field rating methods. Aggregate plot disease levels as estimated by the pipeline correlated highly with field and manually annotated image disease scores; however disease assessments on individual spikes were influenced by field location. The pipeline was able to quantify FHB from images taken with different camera orientations than the original training data, which demonstrates strong generalizability. This innovative pipeline represents a breakthrough in FHB phenotyping, offering precise and efficient assessment of FHB on both individual spikes and plot aggregates. The pipeline is robust across different environments and the potential to standardize disease evaluation methods across the research groups make it a valuable tool for studying and managing this economically significant fungal disease.