IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring", Hyderabad, India, 2002 
  
  
  
     
  
Ba SEE ERE e TBE i 
BR Er eR rt 
NL 
prop Ber er. po 
      
  
  
»* 99 9009459 4n o opi |4 
/ 
"41 48 A FEY 
nm ; 2 Ay inn EA | 
   
  
Figure 4. High-resolution weed map (400 mm/pixel) generated 
with ground truth machine (dark points represent high weed 
density) overlaid on top of soil type map. 
Correlation between weed density and 
Image indices 
  
Resolution (m/pixel) 
  
[>= NNIR — NR —NG — NDVI —- SRI] 
  
Figure 6. Variation of linear correlation between image 
indices and weed density at different resolutions for the 
experiment field. 
Precision chemical application experiments with both smart 
sprayer and remote sensing based map-driven system have shown 
promising results. In Table 2, experiment variable application 
rates are listed. 
09 T 
  
0268 
  
  
  
0.15 
Normalized frequency 
0.1 
  
0.05 
  
   
  
o | 
RC RR SS 
AMAN 
Weed leaf area coverage 
Predicted weed density 
o o o o o o o 
o — N w > e o = 
t 
o 
pet 
  
: 
o 
N 
Actual weed density 
Figure 5. Weed density predicted from RS image (Fig 3) vs. 
actual ground truth weed density (Figure 4). 
Figure 8. Example of weed density distribution in the 
experiment field. 
Herbicide savings from comparing on/off and variable rate 
applications with uniform application are illustrated in Table 3, 
where the single (economical) threshold for on/off application 
was set at a weed density of 1% and the variable rate was set to 
four levels as in Table 2. So, the chemical saving with real-time 
sensors based system is between 52% (one threshold) and 71% 
(four levels). With remote sensing system at 4-m/pixel provide 
highest accuracy in mapping the weed infestation areas. If a map- 
driven system applies herbicide at this resolution, the chemical 
saving would be a function of weed distribution patterns in the 
field. The 
results from our simulated the application showed that 69.3 
percent of the field area needs to be sprayed with 1% weed 
density as threshold the data from our experiment field. In other 
words, remote sensing based system has the potential of saving 
30.7 96 herbicide over uniform application on our experiment 
plots. 
  
  
        
   
Hr SR EU A 
Figure 7. Classified image of the soybean field, in Figure 
2 overlaid with a weed density points collected using the 
smart sprayer, and application map for planting date and 
herbicide treatments. The weed density increased from 
light green to dark green. 
   
    
The 
withi 
imag 
coeff 
resol 
0.85) 
were 
to im 
relati 
speci 
optin 
The | 
the c 
resol 
and c 
This 
appli 
meth 
the v 
This 
and 
Spec 
grad 
unde 
parti 
expe 
And 
wee 
Rese 
Briv 
weet 
IEE] 
118- 
Carc 
and 
364- 
Gop 
Witl 
mul 
Urb.