Classification of PolSAR Images by Stacked Random Forests

Ronny Hänsch,Olaf Hellwich

doi:10.3390/ijgi7020074

Abstract

This paper proposes the use of Stacked Random Forests (SRF) for the classification of Polarimetric Synthetic Aperture Radar images. SRF apply several Random Forest instances in a sequence where each individual uses the class estimate of its predecessor as an additional feature. To this aim, the internal node tests are designed to work not only directly on the complex-valued image data, but also on spatially varying probability distributions and thus allow a seamless integration of RFs within the stacking framework. Experimental results show that the classification performance is consistently improved by the proposed approach, i.e., the achieved accuracy is increased by 4 % and 7 % for one fully- and one dual-polarimetric dataset. This increase only comes at the cost of a linear increased training and prediction time, which is rather limited as the method converges quickly.

Highlights

As active air- or space-borne sensor Synthetic Aperture Radar (SAR) transmits microwaves and records the backscattered echo
Overall, the classification improved significantly, which is illustrated in Figure 3d, which shows the semantic map obtained by the Random Forests (RFs) of the last level and states the final output of the proposed method
This paper proposes using RFs within the stacking meta-learning framework for the classification of Polarimetric SAR (PolSAR) data

Summary

Introduction

As active air- or space-borne sensor Synthetic Aperture Radar (SAR) transmits microwaves and records the backscattered echo. It is independent of daylight, only marginally influenced by weather conditions, and is able to penetrate clouds, dust, and to some degree and, depending on the used wavelength, even vegetation. Those unique properties render it complementary to optical and hyperspectral sensors. There are many modern sensors that acquire PolSAR data, i.e., images that contain complex-valued vectors in each pixel (see Section 2)

Methods

Results

Conclusion