Most crops are cultivated in rows in a defined sowing pattern, i.e. with a constant inter-plant distance. This is an important feature that can be used for crop/weed discrimination. We present two novel methods, one pixel-based and one plant-based, for crop identification taking advantage of the knowledge of the sowing pattern. The pixel-based method uses a lateral histogram of plant pixels along the row direction. The lateral histogram forms a signal with a frequency that corresponds to the distance between crops. The plant-based method first segments out all plants and then uses the location of each plant as feature to find the crops among the weeds. The methods were tested on a set of 143 colour images each covering 80 cm of the sugar beet row. For the pixel-based and the plantbased method, 92% and 96% of the crops were found respectively. The plant-based method has been implemented on a weeding-robot and tested under field conditions. The method is sufficiently fast and robust for real-time control of an intrarow weed-tool performing intra-row cultivation, able to identify 99% of the crops and remove about half of the intra-row weeds. It may be concluded that the methods are well suited for discrimination. However, to be able to recognize and remove a larger amount of weed, the methods need to be completed by recognition methods based on particular features of individual plants such as colour and shape. The weed removal device also needs to be further developed. The advantage of using context information is that the method is not restricted to a specific crop. The most crucial parameters for successful discrimination through our methods are crop upgrowth and weed pressure. For moderate weed pressure (ca. 50 weeds/m) and moderate upgrowth (ca. 70 %), as in our case, context information can be used as an important feature for crop/weed discrimination.
[1]
Johan Fredriksson,et al.
Classification of sugarbeet plants based on contextual information
,
1999
.
[2]
J. Marchant,et al.
Segmentation of row crop plants from weeds using colour and morphology
,
2003
.
[3]
Jing Li Wang,et al.
Color image segmentation: advances and prospects
,
2001,
Pattern Recognit..
[4]
Björn Åstrand.
Mechatronics in agriculture : row guidance system for agricultural field machinery
,
2000
.
[5]
H. J. Olsen.
Determination of row position in small-grain crops by analysis of video images
,
1995
.
[6]
Hans W. Griepentrog,et al.
Individual plant care in cropping systems
,
2003
.
[7]
G. van Straten,et al.
Intra-Row Weed Control: A Mechatronics Approach
,
1998
.
[8]
T. Hague,et al.
A field assessment of a potential method for weed and crop mapping on the basis of crop planting geometry
,
2001
.