Comparison of Different Classification Approaches for Land Cover Classification using Multispectral and Fusion Satellite Data: A Case Study in Ören Forest Planning Unit

Abstract

In this study, the success of different satellite images and classification approaches in land cover (LC) classification were compared. A total of six satellite images, including two passive (Landsat 8 OLI (L8) and Sentinel-2 (S2)) satellite images and four fused satellite images from active (Sentinel-1(S1)-VH and VV polarization) and passive satellite images (L8-S1-VH, L8-S1-VV, S2-S1-VH and S2-S1-VV) were used in the classification in the study. For this purpose, L8, S2, L8-S1-VH, L8-S1-VV, S2-S1-VH and S2-S1-VV satellite images were classified according to three ((Maximum Likelihood Classification (MLC), Support Vector Machine (SVM) and Artificial Neural Networks (ANN)) different image classification approaches using the forest cover types map as gorund data. The results obtained from classification methods were evaluated based on overall accuracies (OA) and kappa coefficients (KC). When the classification successes obtained from the three classification methods are evaluated, it was observed that the KC ranged from 0.66 to 0.95 and the OA ranged from 76.82% to 96.67. The results indicated that the highest OA was displayed by MLC (ranged 85.33% to 96.67%), closely followed by SVM (ranged 80.11% to 91.93%), and finally ANN (ranged 76.82% to 89.92%). In addition, a comparison of classification performance using three utilized classification algorithms was performed. The S1-VH; S1-VV and, S2 and L8 fused images classified with an MLC algorithm produce the most accurate LC map, indicating an OA of 92.00%, 94.00%, 96.67%, 93.33% and a KC of 0.90, 0.93, 0.95, 0.92 for S2 and L8, respectively. Thus, it can be concluded that fused of satellite images improve the accuracies of LC classification.

Evaluating Land Use Land Cover Classification Based on Machine Learning Algorithms

“…However, the amount of land covered by vegetation differs significantly throughout the three categorization techniques. Comparable with many researchers such as [23], [25]. Furthermore, not all machine-learning techniques produce a high-precision LULC map because good results depend on the machine-learning model set-up, training samples, and input parameters.…”

“…Using a map of forest cover categories as ground data, three different image categorization methods, ANN, SVM, and MLC, were used to classify GÜNLÜ satellite images [25]. Categorization technique outcomes were assessed depending on overall accuracies (OA) and kappa coefficients (KC).…”

Beyond Traditional Methods: Innovative Integration of LISS IV and Sentinel 2A Imagery for Unparalleled Insight into Himalayan Ibex Habitat Suitability

“…;Foody, 2002;Keshtkar et al, 2017;Gumma et al, 2019).The Maximum Likelihood (ML) classi er is a commonly employed parametric statistical and supervised classi cation technique in the eld of remote sensing(Jia et al, 2011). The employed strategy is a statistical classi cation method that utilises an average, variance, and covariance approach, taking into account the variable values(Günlü, 2021).The Gaussian mixture model (GMM) is a machine learning approach commonly employed for the purpose of clustering data into distinct groups by leveraging the underlying probability distribution. GMM employed for the classi cation of satellite pictures based on a probabilistic framework(Gu et al, 2007;Chellappa et al, 2009;Okwuashi et al, 2011;Lakshmi et al, 2015).…”

“…structural risk minimization and statistical learning theory(Günlü, 2021). SVM are frequently employed in the context of classi cation tasks and may also effectively handle regression-related issues.…”

“…In the present study, we utilized ve different supervised classi cation algorithms, including Maximum Likelihood, K-nearest neighbour, Support Vector Machine, Gaussian mixed model, and Random Forest. The mentioned classi er algorithms have been widely used in the eld of remote sensing applications and have shown varying levels of accuracy(Jia et al, 2011;Rodriguez-Galiano et al, 2012b;Günlü, 2021). However, the appropriate algorithm selection should be based on the characteristics of the data and study area is crucial for achieving reliable land cover maps.…”