Comparison of Back Propagation Neural Network and Maximum Likelihood classification method in mapping paddy field and sugarcane using multitemporal data of landsat ETM+.

Abstract

The main objectives of this research are to map the paddy field dan sugarcane area with Maximum Likelihood Classification (MLC) and Back Propagation Neural Network (BPNN) methods, and to compare the classification result generated from both classification methods. This research compared parametric method (MLC) and non parametric method (BPNN) by using the same images of Landsat ETM+ and the same training area. Seven bands (multispectral band 1,2,3,4,5,7 and Pancromatic Band 8) from multi temporal Landsat images used as input data for both classification methods. Before Landsat ETM+ used for classification process, it was corrected geometrically and radiometrically. Classification process of both methods (MLC and BPNN) used multistage approach. Landuse classified into 4 classes paddy field, sugarcane, possible paddy field/sugarcane and not paddy field/sugarcane. The target of training area was done based field data. The result show that BPNN classification method has overall accuracy 84,30 % and kappa accuracy 0,64, which is paddy field area 38.040 ha and sugarcane area 5.525 ha. Meanwhile MLC method show overall accuracy 83,26 % and kappa accuracy 0.60 with paddy field area 38.416 ha and sugarcane 6.593 ha. This research is also showed that BPNN get a better accuracy compare to MLC, but the paddy field area generating from both methods is not significantly different.

Perolehan informasi tematik dari citra penginderaan jauh dapat dilakukan dengan proses klasifikasi, yang secara umum dibedakan dalam klasifikasi terawasi (supervised classification) dan tak terawasi (unsupervised classification). Klasifikasi terawasi dapat dibedakan menjadi 2, yaitu klasifikasi parametrik dan klasifikasi non parametrik. Perbedaan antara klasifikasi parametrik dan non parametrik adalah persyaratan statistik distribusi normal, dimana klasifikasi parametrik memerlukan informasi parameter statistik dari daerah sampel pelatihan yang terdistribusi normal, sedang klasifikasi non parametrik tidak mensyaratkan distribusi normal. Metode klasifikasi citra digital yang digunakan dalam penelitian ini adalah maximum likelihood classification/MLC (parametrik) dan metode back propagation neural network/BPNN (non parametrik).