Application of support vector machine for classification of multispectral data

Abstract

Abstract. In this paper, support vector machine (SVM) is used to classify satellite remotely sensed multispectral data. The data are recorded from a Landsat-5 TM satellite with resolution of 30x30m. SVM finds the optimal separating hyperplane between classes by focusing on the training cases. The study area of Klang Valley has more than 10 land covers and classification using SVM has been done successfully without any pixel being unclassified. The training area is determined carefully by visual interpretation and with the aid of the reference map of the study area. The result obtained is then analysed for the accuracy and visual performance. Accuracy assessment is done by determination and discussion of Kappa coefficient value, overall and producer accuracy for each class (in pixels and percentage). While, visual analysis is done by comparing the classification data with the reference map. Overall the study shows that SVM is able to classify the land covers within the study area with a high accuracy. IntroductionLand cover classification is one of the most important remote sensing applications. It has been widely used in many fields such as town planning, studies of environmental change, land resource planning, and geological mapping. Generally, a good classifier should be able to classify pixels into desirable land covers. The factors taken into consideration when selecting a classification method include accuracy, speed and practicality. Among the frequently used methods in classification are maximum likelihood classification (MLC) and artificial neural network (ANN). However, there are drawbacks to these classifications; ANN has been associated with over fitting and local minima problems [5], while MLC needs large training area and assumption that the data are normally distributed. In recent years, there have been an effort to develop better reliable classification methods; support vector machine (SVM) is one of them [10]. SVM is characterised by an efficient hyperplane searching technique that uses minimal training area and therefore consumes less processing time. The method is able to avoid over fitting problem and requires no assumption on data type. Although non-parametric, the method is capable of developing efficient decision boundaries and therefore can minimise misclassification. This is done through finding of optimal separating hyperplanes between classes by focusing on the training cases (support vectors) that lie at the edge of the class distributions, with the other training cases being excluded [13]. This study aims to carry out SVM classification on land covers over Selangor, Malaysia.Originally, SVM is a binary classifier that works by identifying the optimal hyperplane and correctly divides the data points into two classes. There will be an infinite number of hypeplanes and SVM will select the hyperplane with maximum margin. The margin indicates the distance between the classifier and the training points (support vector). Figure 1 illustrates the basic idea of support vector machine [13]. A n...

Discriminating native and plantation forests in a Landsat time-series for land use policy design

“…These results support previous studies that have demonstrated the potential of SVMs to identify different crop (Bahari, Ahmad, & Aboobaider, 2014;Devadas et al, 2012;Löw et al, 2013;Zheng et al, 2015). Sand dune classification accuracy decreased significantly over time.…”

SIFT, SURF, Gabor and Fused Feature Classification Using SVM for Multispectral Satellite Image Retrieval

“…Fig. 4 explains the fundamental concept of support vector machine [4].Many techniques can be implemented to develop the classifier from binary to multiclass i.e. one against all and one against one [9].…”

Monitoring of Land Use/Land Cover and Socioeconomic Changes in South China over the Last Three Decades Using Landsat and Nighttime Light Data

“…Parametric statistical learning techniques such as the maximum likelihood classifier (MLC) fail due to an inability to resolve the interclass confusion. This limitation can be overcome by applying a non-parametric classifier such as a support vector machine (SVM), which does not depend on any assumptions of the class distributions of data [41][42][43]. The SVM is an advanced machine learning algorithm, binary classifier, and a relatively new supervised classification technique [44].…”

“…The SVM is an advanced machine learning algorithm, binary classifier, and a relatively new supervised classification technique [44]. The SVM outperforms the other methods due to its robustness, high classification accuracy, and effective output results, even when using a small training sample [29,[41][42][43]. It operates on the principle of structural risk minimization (SRM) [45] and has overcome the problem of overfitting [41,44].…”