Comparison of vision-based and manual weed mapping in sugar beet

By spraying only strongly weed-infested parts of agricultural fields, the herbicide costs for farmers and the environmental pollution could be reduced. A weed mapping is necessary to obtained information about the actual weed density and distribution on the field. As manual mapping is too much time consuming, a semi-automatic and an automatic weed-mapping method based on image processing were developed and compared to the manual method. Therefore, images were taken under natural field conditions (without additional illumination) on sugar beet fields (76 ha). A feature-based plant discrimination algorithm that calculated different shape features to separate monocotyledonous and dicotyledonous plants based on these images was developed. To validate the developed image analysis system, test images were used; 98.6% of dicotyledonous and 75.0% of monocotyledonous plants were identified correctly.