Classification of Multisensor and Multiresolution Remote Sensing Images Through Hierarchical Markov Random Fields

Abstract

This letter proposes two methods for the supervised classification of multisensor optical and SAR images with possibly different spatial resolutions. Both methods are formulated within a unique framework based on hierarchical Markov random fields. Distinct quad-trees associated with the individual information sources are defined to jointly address multisensor, multiresolution, and possibly multifrequency fusion, and are integrated with finite mixture models and the marginal posterior mode criterion. Experimental validation is conducted with Pléiades, COSMO-SkyMed, RADARSAT-2, and GeoEye-1 data.

Joint Classification of Multiresolution and Multisensor Data Using a Multiscale Markov Mesh Model

“…In this paper we address the problem of data fusion for classification purposes, in the context of different types of very high resolution (VHR) remotely sensed images by proposing a hierarchical hidden Markov model. The proposed methodology, which partly extends the previous methods in [1] and [2], aims at addressing the challenging task of jointly classifying data taken by different sensors at different spatial resolutions, while maintaining an appropriate trade-off between accuracy and computation time. On one hand, this is a typical scenario from an application-oriented perspective, owing to the potential offered by the current space missions with optical (e.g., Pléiades, WorldView-3, SPOT-6/7) and synthetic aperture radar (SAR; e.g., COSMO-SkyMed, TerraSAR-X, RADARSAT-2) VHR payloads in terms of data acquisition capabilities.…”

“…On one hand, this is a typical scenario from an application-oriented perspective, owing to the potential offered by the current space missions with optical (e.g., Pléiades, WorldView-3, SPOT-6/7) and synthetic aperture radar (SAR; e.g., COSMO-SkyMed, TerraSAR-X, RADARSAT-2) VHR payloads in terms of data acquisition capabilities. On the other hand, it is a scarcely explored and difficult problem because of the substantially heterogeneous statistics of the data collected by different sensors (e.g., optical and SAR) and of the need to capture the different spatial details associated with their resolutions [1]. In this paper, we leverage on the modeling power of causal probabilistic graphical models to address this task.…”

“…On the contrary, hierarchical MRFs, whose topology is based on multiscale graphs such as quadtrees, exhibit causality properties that make highly efficient inference procedures feasible [4]. MRFs based on systems of quadtrees have also been found effective in [1] to jointly classify multisensor and multiresolution remote sensing data. However, a common limitation of a hierarchical MRF on quadtree is that, while it captures relations among data at different scales through a Markov chain, it generally does not explicitly characterize spatial dependences within the layer at each scale [4].…”

A Markovian Approach to Unsupervised Change Detection with Multiresolution and Multimodality SAR Data

“…The latter extension would further improve the capabilities of the method to provide change maps characterized by an efficient use of the spatial information in terms not only of local context and multiresolution structure, but also of homogeneous regions [40,56]. Finally, it would be of particular interest to exploit the fusion capabilities of the developed method by incorporating not only multiresolution and multimodality SAR, but also multispectral optical imagery through further appropriate energy contributions [57] or hierarchical graphs [58].…”

Fusing of Optical and Synthetic Aperture Radar (Sar) Remote Sensing Data: A Systematic Literature Review (Slr)

“…(Ahmed et al 2013) used Vegetation indices (optical),Global Environment Monitoring Index GEMI ,Purified Adjusted Vegetation Index PAVI and polarimetric indices SAR (CPR, HV/HH and HV/VV) to detect the subsurface hotspots. The third part of pixel level is two articles based on Hierarchical Markov Random Fields models (Hedhli et al 2015(Hedhli et al , 2017, and finally nine papers applied others different methods including layer stacking (Sameen et al 2016), Genetic algorithm image fusion technique (Ahmed et al 2016), multi-scale decomposition and sparse representation (Zhouping 2015), the combination method band 3, band 7 of Landsat ETM+ with a modified HH polarization of SAR image (Xiao et al 2014) , Closest Spectral Fit (CSF) algorithm with the synergistic application of multi-spectral satellite images and multi-frequency Synthetic Aperture Radar (SAR) data. (Eckardt et al 2013), applied learning Artificial Neural Network at pixel level ANN (Piscini et al 2017), these three typical manifold learning ; ISOMAP, Local Linear Embedding (LLE), principle component analysis (PCA) and two papers the first are not clear and the last without fusion method.…”