Fig. 3.

Summary of the image processing workflow from collection to interpretation. Images are collected and processed with the CNN analysis path, with a subset of images being used for validation data using the manual analysis path. Numbered boxes refer to the steps in the text in the ‘Processing minirhizotron imagery for root trait timeseries’ section. (1) CNN training and segmentation, (2) processing segmented binary images, (3) validation, and (4) post-processing for time series. If validation was poor in Step 3, we continued to train the CNN.