Barrett, Rachel 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Predicted 
BA | OA | TR |WH) ON | CS | PE | BE | PO | PU | BR | CA | PP | PY 
BA 1 0 0 0 2 0 0 0 0 0 0 0 0 0 3 33.3 
OA 0 2 0 0 0 0 0 0 1 0 0 0 0 0 3 66.7 
TR 0 1 3 6 0 0 0 0 0 0 0 0 0 2 12 25.0 
WH | 0 0 1 2 0 0 0 0 0 0 0 0 1 0 4 50.0) 
ON 010/07 01 20[171 011 01070 021 0:18 110229 17 69.005 
CS 0 0 0 0 3 0 1 0 1 0 0 0 2 1 8 0.0 5 
= PE 0 0 0 0 0 1 39 1 1 1 0 1 4 0 48 81.3 8 
= BE 0 0 0 0 0 2 1 2 0 0 0 0 2 0 7 28.6 À 
PO 0 0 1 0 0 0 3 0 17 0 0 1 5 1 28 60.7] m 
PU 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.0 2 
BR 0 0 0 0 1 0 3 0 0 0 0 0 3 0 7 0.0 
CA 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0.0 
PP 0 0 0 0 3 2 0 0 0 0 0 0 52 2 59 88.1 
PY 0 0 0 0 0 1 2 0 0 0 0 0 2 38 | 43 88.4 
1 3 5 8 29 7 50 3 20 1 0 2 79 | 44 | 176 
100.0| 66.7 | 60.0 | 25.0 | 69. | 0.0 | 78.0 | 66.7 | 85.0 | 0.0 | 0.0 | 0.0 | 65.8 | 86.4 | 252 252 
Producers Accuracy (%) 252 | 69.8% 
  
  
  
  
  
  
  
Table 2. Omission/comission matrix for North West Coast 1998/99 growing season 
4 FUTURE WORK 
The objective of this project, to develop a methodology for the systematic recognition of individual crops, has been 
partially achieved. In order to better utilise the technology in a commercial environment, the methodology will be 
tailored individually to five commercially significant horticultural crops. Poppies, pyrethrum, potatoes, peas and onions 
have been selected, not only for their commercial significance but also because, in the initial project, the predictive 
accuracy of these particular crops was the highest of 17 different crops investigated. 
Investigation of the timing of image acquisition, and the number of images required for specific crop growth cycles will 
theoretically enable accuracy to be increased. 
Determination of the most appropriate classifier type for individual crops will enable the most accurate and specific 
prediction methodologies to be documented. 
Given the differences between the SPOT and Landsat platform sensors, resolution and bandwidths, examination of the 
most appropriate combination of SPOT and/or Landsat TM images will determine if a particular platform or 
combination is more suited to a particular crop type. Again, this will enable the use of remote sensing data for crop 
recognition to become more cost effective and accurate. 
5 | CONCLUSION 
Favourable predictive outcomes have been achieved for crop types for which the greatest quantity of both training and 
test data were acquired. These were poppies, pyrethrum and potatoes. This is a logical conclusion from both an 
agronomic and computational point of view, in that the probability of achieving a correct outcome will be enhanced if a 
representative range of horticultural and environmental conditions have been used in both the training and test data. The 
predictive accuracy should be greatly improved by specifically tailoring interpretative images to cropping sequences. A 
point reinforced by the low predictive accuracy with respect to bean crops. 
ACKNOWLEDGEMENTS 
The authors wish to thank the Horticultural Research and Development Corporation (HRDC), Australia and Botanical 
Resources Australia (BRA) for funding this research. 
  
138 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 
Ac fa o 
al ui AME I,