Classification of Land Use and Land Cover in the Brazilian Amazon using Fuzzy Multilayer Perceptrons

Classification of Land Use and Land Cover in the Brazilian Amazon using Fuzzy Multilayer Perceptrons

Toni Pimentel (Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil), Fernando M. Ramos (Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil) and Sandra Sandri (Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil)
Copyright: © 2015 |Pages: 15
DOI: 10.4018/ijncr.2015010104
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

Here the authors propose the use of Fuzzy Multilayer Perceptrons for classification of land use and land cover patterns in the Brazilian Amazon, using time series of vegetation index, taken from NASA's MODIS (Moderate Resolution Imaging Spectroradiometer) sensor. In addition to the traditional Multilayer Perceptron (MLP), three fuzzy implementations were investigated. These methods were applied to a study area of approximately 10.5 km2 on the east of the state of Mato Grosso in the Brazilian Amazon. For validation purposes, the authors compared the best implementation results with the ones given for the same region by the TerraClass 2010 project. The authors observed that our fuzzy MLP correctly classified 81% of the pixels analyzed.
Article Preview

Introduction

The Amazon forest provides invaluable climatic and ecological services. Covering more than 400 million hectares, it provides habitat for innumerous terrestrial plant and animal species. The Amazon region has also a significant influence on the Earth’s climate, both as carbon sink and as a source of atmospheric carbon dioxide. In spite of its relevance, this ecosystem is under threat, having lost more than 15 million hectares of pristine forests since 2000 (Hansen et al., 2013). The causes of land cover change in the Amazon are complex and variable, and involve, among others, permanent cultivation, livestock development, commercial wood extraction, mining, and the extension of overland energy and transport infrastructure (Nepstad et al., 2002). Tracking and understanding the trends and patterns of these land use changes is crucial for the creation and implementation of effective policies for sustainable development and environment protection.

Brazil has the longest tropical forest-monitoring program in the world. Since the late 1980s, Brazil has been collecting remotely sensed information on the Amazon state, and carrying detailed deforestation surveys. Data comes from different sources, including TM/Landsat images and MODIS (Moderate Resolution Imaging Spectroradiometer) products from NASA’s TERRA and ACQUA satellites.

Here we propose the use of Fuzzy Multilayer Perceptrons (MLP) for classification of land use and land cover patterns in the Brazilian Amazon, using time series of MODIS Normalized Difference Vegetation Index (NDVI) (Freitas et al., 2011). We compare the basic MLP algorithm with three extensions, one involving fuzzy concepts (Pal & Mitra, 1992), another involving the notion of ambiguity (Canuto, 2001) and the last involving both. The study area comprises approximately 10.5 km2 at the east of the state of Mato Grosso, Brazil. Following Maus (2013), we used TerraClass in situ classification data to validate our results.

The multilayer perceptron (MLP), using the backpropagation supervised learning algorithm, is the most common neural network found in the literature (Mehrotra, Mohan, & Ranka, 1996), and has been used in a wide range of applications such as pattern recognition (Dimla & Lister, 2000; Jeong, Kim H., Kim D.S., & Lee, 2000; Zhang, 1999), medical i­­mage analysis (Guler, Sankur, Kahya, & Raudys, 1998; Sheppard et al., 1999) and forecasting (Indro, Jiang, Patuwo, & Zhang, 1999). There is a vast literature on the combination of fuzzy set theory (Zadeh, 1965) and the multilayer perceptron (Haykin, 2001; Mehrotra, Mohan, & Ranka, 1996), such as (Stoeva & Nikov, 2000; Pal & Mitra, 1992; Keller & Hunt, 1985).

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 6: 2 Issues (2017): 1 Released, 1 Forthcoming
Volume 5: 4 Issues (2015)
Volume 4: 4 Issues (2014)
Volume 3: 4 Issues (2012)
Volume 2: 4 Issues (2011)
Volume 1: 4 Issues (2010)
View Complete Journal Contents Listing