A New Tree-Based Classifier for Satellite Images

A New Tree-Based Classifier for Satellite Images

Reshu Agarwal (Acadia University, Canada) and Pritam Ranjan (Acadia University, Canada)
Copyright: © 2014 |Pages: 9
DOI: 10.4018/978-1-4666-5202-6.ch003

Chapter Preview



Classified maps play an important role in numerous remote sensing data applications, for example, land-cover change, forest degradation, hydrological modeling, wildlife habitat modeling, and biodiversity conservation. One of the most important products of a raw image is the classified map which labels the image pixels in meaningful classes. Several classifiers have been developed (e.g., Franklin, Peddle, Dechka & Stenhouse, 2002; Pal & Mather, 2003; Gallego, 2004) and implemented worldwide in software packages (e.g., ERDAS IMAGINE, ENVI, IDRISI and ArcGIS) for classifying satellite images. However, accurate prediction of pixel-wise class labels is still a challenge (Blinn, 2005; Song, Duan & Jiang, 2012).

Among various classification methods, Maximum Likelihood (ML) classifier is the most widely used classifier because of its simplicity and availability in image processing softwares (Peddle, 1993). ML classifier is based on a parametric model that assumes normally distributed data, which is often violated in complex landscape satellite images (Lu & Weng, 2007). Non-parametric classifiers do not require stringent model assumptions like normality and gained much popularity. For instance, the classifiers based on k-nearest neighbor (k-NN), artificial neural network (ANN), decision trees and support vector machines (SVM) have shown better performance as compared to ML classifiers (Zhang & Wang, 2003; Bazi & Melgani, 2006; Li, Crawford & Jinwen, 2010; Atkinson & Naser, 2010). Comparison of the classifiers has been an active research area in machine learning. For example, Sudha & Bhavani (2012) concluded that SVM is better classifier than k-NN; and Song et al. (2012) demonstrated that SVM and ANN are comparable; however, SVM often performs slightly better than ANN.

Key Terms in this Chapter

Remote Sensing: Remote sensing is a technique of acquiring information about an object using sensors without making any physical contact with the object.

Classification: It is a problem of identifying an appropriate class label for a new pixel (or observation) based on a training set of data whose class labels are known.

Spectral Resolution: It describes the ability of a sensor to define wavelength intervals. The spectral resolution of an image is inversely proportional to its band width.

Ensemble Method: A statistical technique that uses multiple models to obtain better predictive performance as compared to that obtained from any of its members.

Satellite Image: Satellite imagery consists of photographs of Earth or other planets captured by satellites. We used a satellite image from LANDSAT 5 TM sensor which is a Thematic Mapper (TM) of the 5 th satellite in LANDSAT program.

Kappa Coefficient: It is a measure of accuracy which reflects the difference between actual agreement and the agreements expected by chance.

Decision Tree: It is a decision support tool to display an algorithm that uses a tree-like graph or model of decisions and their possible consequences, including chance event outcomes, resource costs and utility.

Complete Chapter List

Search this Book: