Article Preview
TopLiterature Survey
The literature survey for the problem statement was carried out in three different stages. In the first stage, the types of satellite images to be used, and the pre-processing required was explored. Literature in the second stage deals with a comparison of different machine learning classifiers and the parameters involved. In the third stage, validation mechanisms, and use of multiple classifiers was studied.
For the given problem statement, the use of multispectral images rather than hyperspectral images is more suitable. Literature (Ferrato & Forsythe 2013; Starovoitov & Makarau, 2008; Adam et al, 2010) have shown the difference between the two and how the choice of algorithms changes with the change in the source images. Atmosphere plays a major role in the values of the pixels which are reflected from the surface. Atmospheric correction is, hence an important pre-processing step. Dark Object Subtraction algorithms 1-4, and their pro and cons have been discussed by the authors (Song et al, 2001; Themistocleous, K et al, 2008). Correction for removal of atmospheric defects is important when the images are to be compared in a temporal manner. Radiometric and geometric corrections also form an important role in the LANDSAT images. It is important to identify the appropriate level of pre-processing and the removal of shadows and cloud cover (Young et al, 2017; Shrawankar 2016; Starovoitov & Makarau, 2008).