Identifying the clods with traditional image processing is insufficient due to the
irregular shape of the clods and the variety of soil types. In the literature, clod
analysis is often performed after the clods are removed from their place of formation
or damaged due to sectioning, thus limiting the number of similar approaches. The
highly popular Deep Learning algorithms provided the solution. In this study, YOLO
(You Only Look Once) algorithm was used. For its training, image processing, primarily
the Watershed algorithm, is used for which a Python-based program was developed. After
marking the clods and the background with image processing techniques and then augmenting
it manually, the Watershed algorithm searches for the boundaries of the marked clods,
significantly reducing the manual segmentation process that would otherwise require
numerous human hours. Resulting in a methodology than can be used for both on field
and Discrete Element Method simulations.
