Examination Of Soil Surface Clod Fractions Using Image Processing And YOLO Deep Learning

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.