Kaichang Di 
  
  
  
  
  
  
Figure 3. SPOT multi-spectral image for land use classification (resampled) 
For the sake of comparison, only the Bayes method is applied to classify the image at first. The rectified image is 
overlaid with land use data layers, and the training and test areas are interactively selected. And then the image is 
classified into 8 classes, such as water, paddy, irrigated field, dry land, vegetable field, garden, forest and residential 
area. As shown in the confusion matrix (table 1), the overall accuracy is 77.6199%. Water, paddy, irrigated field, 
residential area and vegetable field are classified with high accuracy. The vegetable field is easlydistinguished from 
other green patches because it is lighter than the others are. Dry land, garden, forest are confused seriously and the 
accuracy is 65.5895, 48.913946 and 59.754496 respectively. And some forest shadows are misclassified as waters. 
  
  
  
  
  
  
  
  
  
Real class 
Classified irrigated d vegetable residential 
water paddy x 1d m em garden forest m Sum 
water 3.900 0.003 0.020 0.013 0.002 0.021 2.303 0.535| 6.797 
paddy 0.004 8.496 0.087 0.151 0.141 0.140 0.103 0.71 9.835 
irrigated field 0.003 0.016 10423 0.026 0.012. 0.076. 0.013 0.623| 11.192 
dry land 0.063 0.48 0.172. 1,709 0.361 2.226 . 2.202 1.080} 8.384 
vegetable field 0.001 0.087 0.002 0.114 3.974 0.634 0.435 0.219 5.465 
garden 0.010 0.009 0.002 0.325 0.263 4.422 4.571 0.065| 9.666 
forest 0.214 0.006 0.000 0271 0.045 1.354 15.671 0.642] 18.202 
residential area 0.132 0.039 0.127 0.080 0.049 0.168 0.839 29.024 30.459 
Sum 4.328 9.135 10.834 2.689 4.846 9.041 26.227 32.901 100 
Accuracy (96) 90.113 93.010 96.204 63.580 81.994 48.913 59.754 88.217 
Overall accuracy = 77.6199% Kappa coefficient = 0.7474 
  
  
Table 1. Confusion matrix of Bayes classification 
  
242 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.