Procedures of image analysis for weed recognition in cereals

Abstract

An automatic weed recognition system is a prerequisite for site specific weed control when using weed threshold concepts. Segmentation, a preprocessing step, is applied to two- or three-dimensional images. In the following step, methods of graph matching, angular measurements within a leaf, and template matching were used to distinguish plants. The weed species Galium aparine and Veronica hederifolia were exemplary examined. In graph matching, typical graphs were identified and in template matching, idealized templates were constructed. The results of this exercise were used to reduce the dimension of the search-space. In the process of template matching, detection rates ranged from 72-83 %; a detection rate of about 91 % was found using angle measurements. Recognition procedures in a knowledge-based system enable using these methods in an automatic recognition system for weed recognition.

Keywords

• Bilderkennungsverfahren
• Galium aparine
• Matching-Verfahren
• Precision farming
• Veronica hederifolia
• Winkelmessung
• wissensbasierte Systeme
• image recognition
• knowledge-based methods
• matching procedures
• measuring angles
• precision farming