Information about the weed population in fields is important for determining the optimal herbicides for the fields. A system based on images is presented that can provide support in determining the species and density of the weeds. Firstly, plants are segmented from the soil. Plants that after the segmentation are divided in multiple parts are selected manually and a cost image is created by weighting pixels according to their relationship to plant. This relationship is based on the colours of pixels and the weighting of nearby pixels. This cost map is used to find the optimal route that connects the plant parts. By using a Support Vector Machine with 18 feature descriptors, suggestions on which plants are present in the images are given. The system is currently able to classify 11 plant species with a precision of 93.0%