SVM and OBIA based Comparative Analysis on LANDSAT Multi Temporal Data for Wetland Mapping

Objectives: The Major objective of this study is to delineate the wetlands using the advanced image processing techniques and study the likelihood of these techniques for the mapping the real extent of wetlands on multi-temporal Lands at data. Methods/Statistical Analysis: The object based image analysis is based on the information of image objects rather than individual pixels. eCognition software is used for object oriented image analysis. The development of the objects and subsequent classification is achieved by multi resolution dissection using fuzzy logic approach1. The support vector machine is one of the supervised classification method is achieved by providing training sets. Findings: The study has been done from the year 1997 to 2014 for the changes in wetlands and their corresponding changes have been observed. The decrease in the areal extent of wetlands has been observed which is due to declining in the annual rainfall,population growth, rapid urbanization and industrialization over decades. It is observed that the classification by Object Based Image Analysis has outperformed the classification performed by SVM. The overall accuracy by this techniques is 76.01%, 75.02%, 77.80%, 76.83% for the data 1997, 2006, 2009, 2014 respectively while the accuracy is 94.2%, 93.2% ,93.8%, 93.1% respectively when classification is performed by object based image analysis. The areal extent of wetlands extracted by Support vector machine in Sq.km is 401.87, 381.96, 263.02, and 147.89 while they are 469.77, 406.46, 309.74 and 155.79 when extracted by object based image analysis for the year 1997, 2006, 2009 and 2014 respectively. Application/Improvements: The changes detected in wetlands over years can be used for the analysis of groundwater recharge, ecosystem & species growth etc.