94.5 95 91.4 94.2 55.9 7.41 Users Acc. RESULT AND DISCUSSION

65 The cutting from vector data above used to calculate accuracy level from classification of both methods back propagation neural network and maximum likelihood. On the calculation method of confusion matrix, there are 3 types of accuracy that can be quantified, i.e. overall accuracy, producer’s accuracy and user’s accuracy. Producer’s accuracy is calculated by dividing the number of correctly classified samples by the column total or the class in the confusion matrix. User’s accuracy is calculated by dividing the number of correctly classified samples by the row total for the category in the confusion matrix. Overall calculation can be seen in Table 4.6 and 4.7 In Table 4.5 below, it can be observed that producer’s accuracy and user’s accuracy of both methods. Table 4.5. The Producer’s Accuracy and User’s Accuracy for Back Propagation and Maximum Likelihood Classification Methods Back Propagation Maximum Likelihood Class Prod. Acc User. Acc Prod. Acc User. Acc Tea Garden 61.6 87.9 76.9 70 Settlement 95.9

97.4 94.5 95

Paddy Field 62.9 93.4 36.4 85 Grass

98.8 91.4 94.2 55.9

Forest 96.4 73.3 71.4 76.8 Farm 78.3 41.5 81.2 20.9 Bush 70.8 36.2 60 22.7 Water Body 47.2 11.3

11.1 7.41

Table 4.6. The Confusion Matrix for Back Propagation Neural Network with 4000 Samples and 1000 Iteration REFERENCE DATA Class Tea Garden Settlement Paddy Field Grass Forest Farm Bush Water Row Total pixel Tea Garden 648 44 36 1 6 2 737 Settlement 22 1437 17 1476 Paddy Field 18 6 545 7 10 4 590 Grass 3 4 85 1 93 Forest 343 5 964 3 1315 Farm 5 38 26 54 5 2 130 Bush 11 4 123 2 85 10 235 Water 4 10 102 1 5 11 17 150 Colum Total pixel 1051 1498 866 86 1000 69 120 36 4726 Producers Acc. 61.66 95.93 62.93 98.84 96.40 78.26 70.83 47.22 Users Acc. 87.92 97.36 92.37 91.40 73.31 41.54 36.17 11.33 C L A S S I F I E D D A T A Overall Acc. 81.15 66 67 REFERENCE DATA Class Tea Garden Settlement Paddy Field Grass Forest Farm Bush Water Row Total pixel Tea Garden 809 49 2 285 1 7 1 1154 Settlement 8 1415 60 2 2 1 1488 Paddy Field 18 2 315 1 3 22 9 370 Grass 6 5 37 81 4 9 3 145 Forest 196 1 14 1 714 3 1 930 Farm 4 68 130 1 56 1 8 268 Bush 8 225 1 2 72 9 317 Water 2 7 36 1 4 4 54 Colum Total pixel 1051 1498 866 86 1000 69 120 36 4726 Producers Acc. 76.97 94.46 36.37 94.19 71.40 81.16 60.00

11.11 Users Acc.

70.10 95.09 85.14 55.86 76.77 20.90 22.71 7.41 C L A S S I F I E D D A T A Overall Acc. 73.34 Based on Table 4.5, it is shown that for classification using neural network method, the highest producer’s accuracy percentage was on grass class 98 and the lowest percentage was on water body class 42. Meanwhile, for the classification using maximum likelihood method, the highest producer’s accuracy was on settlement class 94.5 and the lowest accuracy was on water body class 11.1. These both methods back propagation neural network and maximum likelihood can classify grass and settlement well; it is proved that the accuracy percentage is high. Grass class and settlement was categorized as good due to the percentage resulted from these classes is 98.8 and 94.5 subsequently, it is higher then level of overall accuracy from both methods, which are 81.2 for neural network and 73.3 fro maximum likelihood. Meanwhile, for the both methods, the lowest percentage was on water body class, which is 42 for back propagation neural network and 11 for maximum likelihood method. This condition illustrates us that both methods have difficulties in identifying that particular class. Some reason that can cause it, i.e. because of spectral value for water class is almost the same with spectral class of paddy field, moreover if the paddy field is covered by water when the scan process of the sensor. The mistake can also due to thin cloud that cover the water body class, which are not corrected by using radiometric correction. Table 4.5 is also describing user’s accuracy for both methods back propagation neural network and maximum likelihood. For back propagation neural network method, the highest user’s accuracy was on settlement class 97.4, while for maximum likelihood method was also on settlement class 95. This means, for settlement class determination of reference data is almost 68 the same with class determination of classification result. Meanwhile, the lowest user’s accuracy was on water body 7.41 and 11.3 for maximum likelihood and back propagation neural network method. From the percentages above, the determination of water body class sample of classification result is not suitable with the determination of reference data. This condition occurs due to the limitation in identifying water in reference data Ikonos 2001. The obtained water sample just for river area due to it is assured as water spectral value . An error matrix is a very effective way to represent map accuracy in that the individual accuracies of each category are plainly described along with both the error of inclusion commission errors and errors of exclusion omission errors present in the classification Congalton, 1987. A commission error is simply defined as including an area into a category when it does not belong to that category. An omission error is excluding that area from the category in which it truly does belong. Every error is an omission from the correct category and commission to a wrong category. In back propagation neural network method, all classes have omission error, but only on paddy field class from reference there are all classes that was classified tea garden, settlement, grass, forest, farm, bush and water. Based on the reference, an area is expressed filled by a whole paddy field, but the fact from the class cutting of classification result of neural network, there area 44 pixels was expressed as tea garden, 17 pixels as settlement, 4 pixels as grass, 5 pixel as forest, 26 pixel as farm, 123 pixel as bush and 102 pixels as water body. This condition illustrated that back propagation has encountered difficulties in differentiating paddy field and other classes. Whereas, in maximum likelihood 69 70 method, omission error was also found in almost all category of classes, but paddy field and tea garden class have more level of variation. Based on producer’s accuracy, paddy field has accuracy that can be classified as low under overall accuracy compared to tea garden. So, with the variation of class category that is not included in paddy class, then it will give contribution to the level of its accuracy. For both method back propagation neural network and maximum likelihood, all category classes have commission error, but none of them taken from classification result contains all categories of other classes. Water class has commission error as follows: 2 pixel that supposed to be tea garden, 7 pixel in the settlement, 36 pixel in the paddy field, 1 pixel as farm, 4 pixels as bush and 4 pixels as water body. This condition will give effect to the classification result, which is supposed to be in water body class 4 pixels but it is identified as paddy field 36 pixels. This error occurred for both methods maximum likelihood and back propagation neural network. Error omission diagram can be seen in this Figure 4.17 below. 200 400 600 800 1000 1200 1400 Number of Pixels Tea Garden Settlement Paddy Field Grass Forest Farm Bush Water Reference Ommision Errors Tea Garden Settlement Paddy Field Grass Forest Farm Bush Water Figure 4.17. Ommission Error for Back Propagation Neural Network Method 71 72 200 400 600 800 1000 1200 1400 Number of Pixels Tea Garden Settlement Paddy Field Grass Forest Farm Bush Water Reference Ommision Errors Tea Garden Settlement Paddy Field Grass Forest Farm Bush Water Figure 4.18. Ommission Error for Maximum Likelihood Method

4.4 Comparison Back Propagation Neural Network and Maximum