A Novel Method for Land Cover Classification of Multispectral Satellite Images

Abstract

ABSTRACT newlineToday, Land use and land cover mapping has an inordinate impact in scientific and technological development. The increase in population places a heavy burden on limited natural resources due to the increasing demand on them. The option of identifying natural resources is a possible path to sustainable development. To pursue this path, one requires information on aspects like, cultivable land, barren land, etc, which can be obtained through remote sensing at local and global scale. Remotely sensed satellite images provide a synoptic overview of the whole area in a very short time span. The identification of different land classes from remote sensing or satellite images is a challenging task due to the multispectral bands present in those images. In this regard, Land cover classification is a challenging task in the area of remote sensing and emerges vast applications such as target tracking, change detection and monitoring environmental issues like land reclamation, forestry, disaster management, water quality and planning economic development etc. besides providing broad-scale tracking of natural resources. Therefore, reliable and cost effective technologies to generate such an information system all over the world become apparent. Hence the research goes on finding an optimal algorithm in accurate classification of the land classes for further processing and applications. The classification of various classes such as vegetation, water bodies, buildings, barren land could be determined on the basis of texture features of any kind of original satellite image. newlineThe acquisition of multispectral satellite images is a difficult task as they are not easily available. Hence the input benchmark satellite images are acquired through the official website www.usgs.gov where the required land cover satellite images are enormously found. The processing of those satellite images is then carried out through initial preprocessing using Gaussian filter followed by extraction of their texture features. Per pixel classifica