A Soft Computing Approach for Selecting and Combining Spectral Bands

Juan F H Albarracín,Jefersson A Dos Santos,Ricardo Da S Torres,Marina Hirota,Rafael S Oliveira

doi:10.3390/rs12142267

Abstract

We introduce a soft computing approach for automatically selecting and combining indices from remote sensing multispectral images that can be used for classification tasks. The proposed approach is based on a Genetic-Programming (GP) framework, a technique successfully used in a wide variety of optimization problems. Through GP, it is possible to learn indices that maximize the separability of samples from two different classes. Once the indices specialized for all the pairs of classes are obtained, they are used in pixelwise classification tasks. We used the GP-based solution to evaluate complex classification problems, such as those that are related to the discrimination of vegetation types within and between tropical biomes. Using time series defined in terms of the learned spectral indices, we show that the GP framework leads to superior results than other indices that are used to discriminate and classify tropical biomes.

Highlights

Remote sensing is important for monitoring and modeling vegetation dynamics and distribution in large-scale ecological studies
The ever growing variety of sensors with different technical specifications, such as spatial/spectral resolutions or acquisition protocols, may lead to low performance of traditional spectral indices, as they may not accurately reflect the same properties related to data obtained from different sources [2]. In all of those situations, the use of techniques that can learn existing patterns of interest from data can lead to more effective indices that are tailored to specific classification-dependent applications. We address those shortcomings by proposing a methodology that employs a Genetic-Programming (GP)-based framework for learning spectral indices that are specialized in discriminating pixels that belong to different classes
We introduced a Genetic-Programming-based framework for learning specialized vegetation indices from data, and tested it for discriminating between tropical South American forests and savannas, as well as the vegetation types existing within them

Summary

Introduction

Remote sensing is important for monitoring and modeling vegetation dynamics and distribution in large-scale ecological studies. Most of the studies applying these techniques use multispectral indices that are based on a ratio (or some other simple mathematical relation) of the reflectance at two or more wavelengths [1,2] These indices allow for complex data analyses, such as better detecting and visualizing specific targets, like vegetation vigour [3,4], water bodies [5], and deforestation patterns [6], since they make explicit complex interactions between the bands that cannot be evidenced individually. In order to determine how fit an individual is to be selected, i.e., how good it solves the problem, a criterion must be defined in terms of a fitness function This function has as input an individual, and outputs a score that allows for comparison of all the individuals in the population. This score determines the chance of individuals to be selected

Results

Discussion

Conclusion