Fusarium head blight (FHB) is an economically damaging fungal pathogen of wheat that can cause yield losses over 50%. Host resistance is one of the most effective approaches for disease control, however time, labor requirements, and human subjectivity limit phenotyping efforts. In this study, a novel, high-throughput phenotyping rover was deployed to capture in-field RGB images of inoculated wheat spikes throughout the 2021 and 2022 growing seasons. A deep neural network pipeline was developed to classify wheat spikes in the complex images, segment healthy and diseased tissue, and calculate the FHB disease percentage as the region of intersection between the spike and disease masks. To validate the pipeline's accuracy, inferences generated by the model at both the plot and spike levels were compared with disease scoring performed by five human raters in the field and manual image analysis. Using the phenotyping rover and our developed FHB quantification pipeline surpassed conventional rating methods in both precision and throughput. Aggregate plot-level disease scores derived from pipeline outputs strongly correlated with disease scores assigned by raters in the field and from image analysis. The pipeline disease annotations on single spike images correlated well to manual image annotations by raters, however there was a locational bias for some images. The FHB pipeline displayed generalizability as it performed well across environments, camera angles, and disease progression. These results represent a significant advancement in FHB phenotyping, affording precise and efficient quantification of disease severity on a spike and plot level at a scale unachievable using current rating methods.