A metaheuristic feature-level fusion strategy in classification of urban area using hyperspectral imagery and LiDAR data

Hadiseh Hasani,Farhad Samadzadegan,Peter Reinartz

doi:10.1080/22797254.2017.1314179

Hadiseh Hasani, Farhad Samadzadegan + Show 1 more

Open Access

https://doi.org/10.1080/22797254.2017.1314179

Copy DOI

Abstract

ABSTRACTOne of the most sophisticated recent data fusions in remote sensing has involved the use of LiDAR and hyperspectral data. Feature-level fusion strategy is applied based on extraction of several recent proposed spectral and structural features from hyperspectral and LiDAR data, respectively. In order to optimize classification performance, feature selection and determination of classifier parameters are carried out simultaneously. Referring to complexity of search space, cuckoo search as a powerful metaheuristic optimization algorithm is applied. Experiments show that the proposed method can improve the overall classification accuracy up to 6% with respect to only hyperspectral imagery. The obtained results show the classification improvement for the tree, residential and commercial classes is about 4%, 21% and 35%, respectively.

Highlights

The interest in the joint use of remote-sensing data from multiple sensors such as aerial imaging, LiDAR, multispectral, hyperspectral and SAR data has been remarkably increased for environmental monitoring and land-use management (Ban & Jacob, 2016; Forzieri, Tanteri, Moser, & Catani, 2013; Latifi, Fassnacht, & Koch, 2012)
An optimum hybrid classification of hyperspectral imagery and LiDAR data based on cuckoo search is proposed
To evaluate the performance of the proposed method, experiments are performed on compact airborne spectrographic imager (CASI) hyperspectral imagery and LiDAR-derived Digital Surface Model (DSM) acquired by the NSFfunded Center for Airborne Laser Mapping, both at the same spatial resolution (2.5 m)

Summary

Introduction

The interest in the joint use of remote-sensing data from multiple sensors such as aerial imaging, LiDAR, multispectral, hyperspectral and SAR data has been remarkably increased for environmental monitoring and land-use management (Ban & Jacob, 2016; Forzieri, Tanteri, Moser, & Catani, 2013; Latifi, Fassnacht, & Koch, 2012). Liu et al (2014) applied particle swarm optimization to determine the SVM kernel and margin parameters in classification of hyperspectral imagery and the results show the superiority of the proposed method in comparison to the grid search method Feature selection is another essential step in classification of high dimension data. LiDAR-derived DSM preprocessing In order to analyze the DSM accurately, nDSM is extracted from DSM by morphological grayscale reconstruction (Arefi & Hahn, 2005) For this purpose, the disk-shaped structuring element (SE) is used and its size is defined based on the object’s size in the dataset. At the first step of the proposed method, an enlarged feature space based on hyperspectral imagery and LiDAR data is generated. Three descriptors consisting of Semi-variogram, Madogram, and Rodogram are computed by Equations (7–9), respectively (Chica-Olmo & AbarcaHernández, 2004)

NðhÞ jZðxiÞ

Experimental results

Conclusion