IMPROVED CLASSIFICATION OF MODERATE RESOLUTION SATELLITE IMAGE USING BAND RATIO, NBAI AND PC TRANSFORM

摘要

This study explored the technique of using band ratio (Band 4/Band 3 of Landsat ETM+), Normalized Built-up Area Index (NBAI) and Principal Component Transform (PC 4) as supplementary bands to aid in properly classifying pixels that are either built-up or soil (the two most confused classes) via the supervised Maximum Likelihood Classifier. The technique was applied to a test area located in the island of Negros, Philippines. The technique led to an improvement of 5.2% to the classification's over-all accuracy as compared to the image classification using only the six original bands of Landsat(l-5, 7). Moreover, the producer's and user's accuracies for built-up and soil classes are improved by a maximum increase of 4.6% and kappa coefficient by 8%. This made the image classification within the suggested minimum level of interpretation accuracy (85%) for land use/cover classes derived from remotely sensed data.