Classification of Multi-spectral Landsat Satellite Imagery Dataset using Feature Extraction Methods

Abstract

The feature extraction of multi-spectral Landsat satellite imagery Dataset is essential for vegetation monitoring, urban planning, change assessment, and other land-use applications. The spatial information provided by Remote sensing satellite imagery data is helpful for planning and decision-making policies. In the present study, classify the features of the multi-spectral Landsat satellite imagery dataset in different periods using the feature extraction method, and is produced the spatial maps of the study area. The study is to analyze the appropriate method of feature extraction for classifying the orchards, vegetation, rangeland, agricultural land, wetland, water body, and urban land using multi-temporal satellite dataset. In this study, use the three feature extraction methods are support vector machine (SVM), minimum distance (MD), and Maximum likelihood classifier (MLC) for supervised pixel-based classification using medium resolution (30 m) satellite dataset. The accuracy of feature extraction method is performed by the MLC (86.29% and 93% in the year 2003 and 2017) and SVM (86.37% and 90% in the year 2003 and 2017). The result of the presented study shows MLC and SVM classifier performs similar results but better than MD classifier for land-use/cover features classification. The classified spatial maps provide the essential spatial information for land-use changes occurred during the last 15 years (2003 to 2017).