IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India, 2002 
Table 1. Summary of weed density data collected using the ground truth machine. The weed density is expressed as a ratio of the area 
covered by weeds to the total ground area. 
  
  
  
  
Weed density 
Field Name Year Count 
Mean Variance Minimum Maximum 
a Plot 1998 3660 0.3193 0.0406 0.0 0.9928 
REBEL Plot. 1998 2310 0.0289 0.0023 0.0 0.4373 
RE-CEL 1998 2897 0.0326 0.0045 0.0 0.3384 
RF-CW 1999 26521 0.3732 0.1301 0.0 1.0000 
RF-DEI 1999 17760 0.2206 0.0878 0.0 1.0000 
2.3 Experiments 
The study was conducted in three soybean fields in the 
Agricultural Engineering Research Farm at the University of 
Illinois at Urbana-Champaign. The CIR images of the fields were 
acquired on a monthly basis during the crop season. However, 
since ground data was collected once during the cropping season, 
the CIR image corresponding to that date was used in this study. 
The crop was planted at a rate of 150,000 plants/acre, and at 75 
cm row spacing in all fields. The fields in this study had several 
weed types such as morning glory, cocklebur, velvet leaf, crab 
grass, giant foxtail, etc. Initially the CIR images were rotated and 
cropped if necessary and then geo-referenced using ground 
control points (GCP) collected at the corners of the experiment 
plots, and the boundary file for each field. The ground truth weed 
map was overlaid with the classified image to visually compare 
the variations in weed density with the spatial variations within 
the image. For this study, five image indices derived from the 
CIR image were selected based on the correlation. between 
various image indices and weed density. The image indices 
selected were NDVI, SRI, and the three raw image bands 
normalized by the total intensity in the color-infrared region. The 
NIR, R, and G bands after intensity normalization were referred 
to as NNIR, NR, and NG respectively. These intensity- 
normalized bands were able to eliminate the variability due to 
different soil types as well as variable illumination in this 
particular study. 
FOV of the 
> = al cameras 
Le mA 
FT £ 5 cour, fone 
PE» 
     
  
   
  
     
  
    
ud 
Figure 2. Spray and mapping system set up (top view). 
For the chemical saving experiment, both ground truth about 
weed infestation and remote sensing weed infestation area data 
were collected at the same time the sprayer field experiment was 
carried 
out. The amount of chemical consumption was recorded for both 
conventional and precision chemical applications. The 
conventional chemical comparison test was conducted with the 
same sprayer. Since the smart sprayer only used a portion of the 
boom on one side of the sprayer, the remaining part of the spray 
boom was set up to simulate conventional uniform rate 
application. The chemical input from the "smart spray" portion is 
recorded as a map (amount vs. GPS location) and conventional 
spray chemical input is then equal to the total amount minus the 
smart sprayer amount. Weed distribution data from the weed map 
were plotted and analyzed with respect to these control zones 
(sampling grids). Weed-free or weed-infestation areas at different 
threshold levels were calculated. The result of this analysis 
demonstrated how different weed control methods affect the 
sprayer's performance. 
Table 2. Proposed variable application rates (VAR) 
  
Weed density (96) 0-1 1-2 2-10 >10 
  
Application 10 33 66 100 
rate (%) 
  
3. RESULTS AND DISCUSSION 
Figure 3 is an example of color near infrared images from the 
experiment field. Figure 4 shows the chemical application map, 
which is also the weed infestation area distribution map. Table 1 
summarized the weed density data collected with ground based 
mapping system. Figure 5 shows an example of the correlation 
between weed maps and processed-aerial image data. The 
correlation between the aerial image and the weed map generated 
with the ground system is found to be a function of the spatial 
resolution of aerial imaging system. High-resolution remote 
  
    
sensin 
m/pixe 
map v 
resolu 
and tl 
should 
weed « 
due t 
referei 
accura 
weed : 
at hi 
correl: 
4.5-m/ 
points 
possib 
image 
GPS vw 
Table 
  
Sing 
thres 
meth 
Vari: 
rate 
meth 
  
.* Stan 
Figur 
of a te 
field,