Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Yield mapping of potatoes with machine vision requires potato mass estimation from information in the images. Several methods for mass estimation exist but are usually based on 3D information. Objective of the project was to develop a method for mass estimation based on 2D information from a line scan camera above a transport belt behind the digger web of a potato harvester. Different dimensions of two varieties potatoes, different shapes and size classes were measured by hand and image processing to develop different models. Mass and volume of individual potatoes were measured by hand too. Regression analysis was used to determine relations between potato volume and potato dimensions. The average prediction error of the best model was 0.27% for Bintje and Agria. This model was used to estimate potato volume from line scan images of harvested potatoes on a moving transport belt. Applying the model to this dynamic situation showed an average deviation on batch level ranging from 1.5 to 2.6%. This gives, together with an estimated error of 2% for potato density, a mass estimation error between 3.5 and 4.6%. In practice cluster forming occurs on the transport belt. Clusters are seen as one big potato if no special measures are taken. Therefore a method was developed to detect clusters and to singularize these clusters by software. Cluster detection and separation performed well. However, some improvements are still necessary for more complex situations. The work described in this paper shows that there are good perspectives for a machine vision based online potato yield mapping system.