Simulated Dataset to Verify the Overlapping and Segregation Problem on Computer Vision Granullometry of Fertilizers

Abstract Fertilizers are an important tool for agriculture to correct the nutrients of the soil. Several analyses are made to guarantee the quality of this product. The particle size analysis indicates how well the fertilizer will penetrate in the soil by the size. The idea is to estimate the size of the grain during the production process. The production of fertilizers is a very complex production involving meters of pipes and conveyor belts where the grains are composed and transformed into the final product. The classic method to estimate the size is the mechanical sieving, an invasive and time-consuming method. A non-invasive and cost-effective method is the digital image processing (DIP) technique applied online in the production flow. In this case, a camera can be localized in the top of a conveyor belt capturing grain images directly during their composition. However, due to the number of grains (tons of grains) present on the image, this method depends on particle separation to avoid the particle segregation and grain overlapping on sampled images. In this work, we investigate how a digital image processing algorithm for particle size analysis of fertilizers is affected by the segregation and overlapping issues. We propose a grain surface simulator to create different scenarios of particle dispersion, a useful tool to speed up the process of creation of data-sets about fertilizers. The results show how the overlapping and segregation of grains influences in the particle size analysis by DIP, and how these interferences in extreme situations could generate biased results.