23 February 2016 Forest encroachment mapping in Baratang Island, India, using maximum likelihood and support vector machine classifiers
Author Affiliations +
J. of Applied Remote Sensing, 10(1), 016016 (2016). doi:10.1117/1.JRS.10.016016
Maximum likelihood classifier (MLC) and support vector machines (SVMs) are commonly used supervised classification methods in remote sensing applications. MLC is a parametric method, whereas SVM is a nonparametric method. In an environmental application, a hybrid scheme is designed to identify forest encroachment (FE) pockets by classifying medium-resolution remote sensing images with SVM, incorporating knowledge-base and GPS readings in the geographical information system. The classification scheme has enabled us to identify small scattered noncontiguous FE pockets supported by ground truthing. On Baratang Island, the detected FE area from the classified thematic map for the year 2003 was ∼202  ha, and for the year 2013, the encroachment was ∼206  ha. While some of the older FE pockets were vacated, new FE pockets appeared in the area. Furthermore, comparisons of different classification results in terms of Z-statistics indicate that linear SVM is superior to MLC, whereas linear and nonlinear SVM are not significantly different. Accuracy assessment shows that SVM-based classification results have higher accuracy than MLC-based results. Statistical accuracy in terms of kappa values achieved for the linear SVM-classified thematic maps for the years 2003 and 2013 is 0.98 and 1.0, respectively.
© 2016 Society of Photo-Optical Instrumentation Engineers (SPIE)
Laxmi Kant Tiwari, Satish K. Sinha, Sameer Saran, Valentyn A. Tolpekin, Penumetcha L. N. Raju, "Forest encroachment mapping in Baratang Island, India, using maximum likelihood and support vector machine classifiers," Journal of Applied Remote Sensing 10(1), 016016 (23 February 2016). https://doi.org/10.1117/1.JRS.10.016016

Remote sensing

Image classification

Geographic information systems

Statistical analysis

Global Positioning System


Accuracy assessment

Back to Top