Change Detection Analysis of Land Cover Features using Support Vector Machine Classifier

Abstract


 
 
 Remote sensing (RS) is crucial for geographical change studies such as vegetation, forestry, agriculture, urbanization, and other land use/land cover (LU/LC) applications. The RS satellite imagery provides crucial geospatial information for observation and analysis of the entire earth's surface. In the proposed study, Multitemporal and multispectral Landsat satellite imagery is used to feature extraction of LU/LC of the Haridwar region. The preprocessing of used imagery is essential for accurately classify the land cover features using image preprocessing methods (geometric correction, atmospheric correction, and image transform). It helps to classify and change detection of land cover features accurately. After preprocessing of imagery, land cover features are divided into seven feature classes using the region of interest (ROI) tool with google earth image and topographic map. The Support vector machine (SVM) is a supervised learning method used to classify the land cover features of the study area. SVM classifier accurately classifies the imagery of the different years 2017, 2010, 2003, and 1996 with 90.00%, 82.75%, 86.37%, and 83.38% accuracy. The post-classification method is used to detect changes in land cover features. From 1996 to 2017, orchards and vegetation are rapidly decreased by 13,698.36 ha and 1,638.81 ha due to unplanned development in urban and industrial areas of the Haridwar region. The resultant LU/LC change information is important for monitoring and analyzing land cover changes of the study area.