Comparison of conventional and advanced classification approaches by Landsat-8 imagery

Abstract

Abstract.Over the last few years, most areas of Iran including Golestan province have posed considerable risks to human use and future development. Therefore, several reliable techniques are required to quantify, monitor and update land use maps of these areas to explore rates of environmental retreats. In this study, Landsat-8 ETM+ imageries of 2013 were consequently processed via Object Based Image Classification (OBIC) as an advanced approach and Maximum Likelihood Classification (MLC) as a conventional pixel based approaches to classify land use in Golestan province in Northeastern of Iran using an ENVI 4.8 and eCognition software packages. Overall accuracy and kappa coefficients were calculated to assess these approaches, which were respectively 94.69% and 0.93 for OBIC and 81.53% and 0.75 for MLC. In addition, accuracy rate over 85% indicates satisfactory for land use mapping and planning purposes. Our result showed that pixel-based classification approach do not lead to sufficiently accurate land use maps; and land use mapping using OBIC approach provides better classification accuracy especially when we have an extensive study area. Thus, development of this approach is appropriate for satellite images with moderate resolution.

Object based burned area mapping with random forest algorithm

“…There were many studies that applied to OBIA to medium and low resolution images for burned area mapping (Gitas, Mitri et al 2004, Polychronaki and Gitas 2012, Katagis, Gitas et al 2014, Kavzoglu, Erdemir et al 2016, (Gitas, Mitri et al 2004). Analyzing the studies using medium resolution satellite images to compare these two approaches, OBIA gives more accurate results than PBIA (Estoque, Murayama, & Akiyama, 2015;Gao, Mas, Kerle, & Pacheco, 2011;Gilbertson, Kemp, & van Niekerk, 2017;Varamesh, Hosseini, & Rahimzadegan, 2017). Also, OBIA reduce the salt and pepper effect that cause misclassified pixel on satellite images (Phiri & Morgenroth, 2017) (Gao et al 2011).…”

Analysis of landscape evolution in a vulnerable coastal area under natural and human pressure

“…For the land cover classification, we adopted the Iterative Self Organizing Data Analysis (ISODATA), one of the most utilized and robust methods in the multispectral and multitempoal unsupervised classification (Tou and Gonzalez 1974;Simoniello et al 2008;Pope and Rees 2014;Varamesh et al 2017). As input for the unsupervised algorithm, for both OLI and TM we used all the VNIR and SWIR bands (30 m) excluding the thermal bands for the low spatial resolution (100 m and 120 m, respectively).…”