Agriculture Reference
In-Depth Information
0
50
100
150
200
250
300
350
400
450
500
550
stone
weed
straw
338 377 416 455 494 534 573 612 651 690 729 769 806 847 886 925
wavelength [nm]
Fig. 10.4
Video spectrometer image with typical refl ections of a stone, weed and straw in the
spectrum of 338-925 nm with soil as background (Sökefeld et al.
2007
)
Figure
10.4
shows an image of a typical video spectrometer scene. Nearly over
the entire analyzed spectrum, stones show a strong refl ection. The refl ection of dead
organic material like straw is very similar to the refl ection of stones. Living plants
refl ect moderately below 570 nm and have a strong refl ection above 690 nm. The
characteristic decrease of refl ection between 610 and 690 nm is typical for green
plants because of the absorption of this band by chlorophyll.
Based on these studies, it was concluded that a normalized difference between
images above 700 nm and images between 610 and 690 nm would result in high
quality images with a strong contrast between green plants and background. A bi-
spectral camera was developed computing differential images of the infrared and
red wavebands (Figs.
10.5
and
10.6
). The resulting images were saved on the hard
disc of a computer for further processing. The
bi-spectral camera
allows real-time
detection and identifi cation of weed species in arable crops. The speed of the image
analysis is high enough to use the cameras for online weed control in combination
with a fi eld sprayer.
In the subsequent text, the processing of the data is dealt with in an abbreviated
and condensed manner. Details to this are explained in the literature cited.
A
circular closing operator
of size fi ve was used to connect most leafs of a
single plant. For the extracted plants, various numerical features were computed
that refl ect the form of the plant species:
•
Region-based: these features are based on the region pixels, which are defi ned as
a connected set of pixels. Examples are the size, compactness, minimum and
maximum diameter and several statistical measures (statistical moments (Jähne
2001
), Hu moments (Hu
1962
))
