COMPARISON OF LAND COVER CLASSIFICATION METHODS OVER MECCA USING ALOS DATA

Four different classifiers were used to analyze multispectral data. This research provides a comparison of Minimum Distance (MD), Maximum Likelihood (ML), Neural Network (NN) and Frequency-based Contextual (FBC) classifiers in the context of land cover mapping in arid environment of Mecca city. MD and ML represent the traditional method whereas NN and FBC represent the advanced method. Classification results were compared in order to evaluate the suitability of the four classification methods in the given study area. A 10m spatial resolution ALOS AVNIR-2 image of Mecca in Saudi Arabia was used for classifying six classes. The classes are urban, mountain, land, vegetation, ritual area and shadow.

A comparative analysis of land cover for all classifiers was done based on three factors which are the overall classification accuracy, performance in the heterogeneous area and the effect of the training size to the classifier. From the experimental result, it was found that the advanced method produces a more accurate result than the traditional method. The overall classification accuracy of NN method was 84.2% with a 0.757 kappa coefficient and FBC had an accuracy of 81.6% with a 0.722 kappa coefficient. The accuracy of the other classifiers ranged from 64.2% to 77.6% with kappa coefficients from 0.479 to 0.649.

The advanced method (NN and FBC) also show its ability in dealing with the difficult challenge of heterogeneous area compare to the traditional method. From the perspective of training size, traditional method needs to have a large number of training data in order to perform classification whereas the advanced method does not depend on the training size for classification.

 

 

 

Empat pengelas yang berbeza telah digunakan untuk menganalisis data multispektrum. Kajian ini menyediakan perbandingan di antara pengelas-pengelas Minimum Distance (MD), Maximum Likelihood (ML), Neural Network (NN) dan Frequency-based Contextual (FBC) dalam konteks pemetaan litupan tanah di persekitaran gersang bandar Mekah. MD dan ML mewakili kaedah tradisional manakala NN dan FBC  mewakili kaedah maju. Hasil pengelasan dibandingkan  untuk menilai kesesuaian empat kaedah klasifikasi di kawasan kajian. Imej Mekah di Arab Saudi beresolusi ruang 10m  daripada ALOS AVNIR-2 telah digunakan untuk mengklasifikasi enam kelas. Kelas-kelas tersebut adalah bandar, gunung, tumbuh-tumbuhan, kawasan ibadah dan bayang-bayang.

Analisis perbandingan litupan tanah untuk semua pengelas  dilakukan berdasarkan tiga faktor iaitu  ketepatan klasifikasi keseluruhan, prestasi di kawasan heterogen dan kesan saiz latihan terhadap pengelas. Daripada hasil eksperimen, didapati bahawa kaedah maju menghasilkan keputusan yang lebih tepat daripada kaedah tradisional. Ketepatan klasifikasi keseluruhan bagi kaedah NN adalah 84.2% dengan 0.757 pekali kappa dan FBC mempunyai ketepatan 81.6% dengan 0.722 pekali kappa. Ketepatan pengelas lain adalah dalam julat 64.2% sehingga 77.6% dengan pekali kappa 0.479 sehingga 0.649.

Kaedah maju (NN dan FBC) juga menunjukkan keupayaannya dalam berusan dengan kesukaran cabaran di kawasan heterogen dibandingkan dengan kaedah traditional. Daripada perspektif saiz latihan, kaedah tradisional memerlukan sejumlah besar data latihan untuk melaksanakan pengelasan manakala kaedah maju tidak bergantung pada saiz latihan untuk pengelasan.