Alternate neural network tools for pattern classification in satellite imageries


Case : Land use / Land cover Classification in Delhi Ridge

Table 1. Confusion Matrix of Neural Network Classifier
Number of hidden neurons 6 Total Accuracy 89%
Number of Training Pixels 252 Verification Set 2000

Class Name Unclassified Dense forest Open forest Degraded forest Built-up Area Forest Blank Accuracy %
Dense forest 22 244 10 29 4 17 74.00
Open forest 14 6 336 1 6 0 92.00
Degraded forest 28 15 5 438 13 6 86.00
Built-up Area 17 1 0 0 617 1 97.00
Forest Blank 13 3 0 0 5 149 87.00


Table 2. Confusion Matrix of Neural Network Classifier
Number of hidden neurons 6 Total Accuracy 89%
Number of Training Pixels 252 Verification Set 1500

Class Name Unclassified Dense
forest		 Open
forest 		Degraded
forest		 Built-up
Area 		Forest
Blank 		Accuracy %
Dense forest 17 193 6 24 2 10 76.00
Open forest 11 4 279 1 5 0 93.00
Degraded forest 21 13 4 296 9 5 85.00
Built-up Area 10 1 0 0 461 0 97.00
Forest Blank 10 3 0 0 2 113 88.00


Table 3. Confusion Matrix of Neural Network Classifier
Number of hidden neurons 6 Total Accuracy 88%
Number of Training Pixels 252 Verification Set 1000

Class Name Unclassified Dense
forest 		Open
forest		 Degraded
forest		 Built-up
Area		 Forest
Blank 		Accuracy %
Dense forest 12 106 1 20 2 7 71.00
Open forest 8 2 151 0 4 0 91.00
Degraded forest 20 8 2 234 6 4 85.00
Built-up Area 5 1 0 0 322 0 98.00
Forest Blank 6 2 0 0 1 76 89.00


Table 4. Confusion Matrix of Neural Network Classifier
Number of hidden neurons 6 Total Accuracy 88%
Number of Training Pixels 252 Verification Set 500

Class Name Unclassified Dense
forest		 Open  
forest 		Degraded
forest 		Built-up
Area		 Forest
Blank		 Accuracy %
Dense forest 6 67 0 8 1 1 80.00
Open forest 5 2 95 0 3 0 90.00
Degraded forest 11 6 1 92 3 1 80.00
Built-up Area 2 1 0 0 157 0 98.00
Forest Blank 3 2 0 0 0 33 86.00



Case: Species Classification in Shimla Region

Table 5. Confusion Matrix of Neural Network Classifier
Number of hidden neurons 8 Total Accuracy 81%
Number of Training Pixels 408 Verification Set 2000

  UNCLASSIFIED A B C D E F ACCURACY % A - Upland Hardwoods mixed with coniferous
B - Exposed rocks, Barren lands
D - Agricultural land and habitation
E - Blue pine (kail)
F - Fir/Sprue
A 19 281 4 20 28 14 8 75
B 14 1 318 1 12 4 0 90
C 40 3 0 190 2 17 4 74
D 21 6 3 4 413 13 2 89
E 48 1 3 4 8 162 0 71
F 5 7 3 19 3 21 274 82


Table 6. Confusion Matrix of Neural Network Classifier
Number of hidden neurons 8 Total Accuracy 81%
Number of Training Pixels 408 Verification Set 1500

  UNCLASSIFIED A B C D E F ACCURACY %
A 12 245 4 15 26 11 7 76
B 9 1 243 1 10 2 0 91
C 27 2 0 105 0 12 4 70
D 17 6 3 3 338 10 1 89
E 31 1 3 3 8 91 0 66
F 2 4 3 13 3 17 207 83


Table 7. Confusion Matrix of Neural Network Classifier
Number of hidden neurons 8 Total Accuracy 81%
Number of Training Pixels 408 Verification Set 1000

  UNCLASSIFIED A B C D E F ACCURACY %
A 6 168 1 8 17 8 5 78
B 7 1 154 1 10 1 0 88
C 17 1 0 77 0 10 1 72
D 13 5 2 2 215 7 1 87
E 22 1 0 3 5 64 0 67
F 2 2 1 9 3 11 139 83


Table 8. Confusion Matrix of Neural Network Classifier
Number of hidden neurons 8 Total Accuracy 81%
Number of Training Pixels 408 Verification Set 500

  UNCLASSIFIED A B C D E F ACCURACY %
A 2 86 1 4 5 5 3 81
B 5 0 77 0 4 1 0 88
C 8 1 0 29 0 4 1 67
D 8 3 2 2 107 36 0 85
E 10 0 0 2 3 4 0 70
F 0 1 1 7 3   72 81


Page 2 of 2
| Previous |