Scattering and contextual features fusion using a complex multi-scale decomposition for polarimetric SAR image classification

Abstract

Polarimetric synthetic aperture radar (PolSAR) images contain rich information about back-scattering and physical characteristics of targets. So, they have high ability for discrimination of different land cover classes. The aim of this research is to introduce an efficient method for PolSAR image classification. Extraction of both scattering and contextual features is important for class discrimination. Therefore, the scattering and contextual feature fusion (SCF) method is proposed to fuse the extracted polarimetric and morphological features through applying a complex multi-scale decomposition. The dual tree complex wavelet transform is used to decompose each scattering feature map into its details and approximate components. The contextual feature maps are decomposed in a similar way. Then, details of two kinds of feature maps are fused region by region. This process is also done for the approximation components containing the low frequency information. The result will be a high dimensional fused feature space. The principal discriminant analysis (PDA) is proposed to reduce the data dimensionality with discarding noisy components and increasing the class discrimination. The extracted features are then fed into a simple classifier to obtain the classification map. Three L-band PolSAR images acquired by airborne synthetic aperture radar (AIRSAR) and electronically steered array radar (ESAR) are used for doing experiments. The SCF method shows superior classification results with respect to several state-of-the-art PolSAR classifiers. For example, for the Flevoland dataset containing 15 classes, without applying post processing, the SCF method results in 95.22% overall accuracy compared to 2DCNN with 91.84% and 3DCNN with 93.94% overall accuracy. With applying post processing, the classification results of SCF, 2DCNN and 3DCNN are increased to 99.55%, 98.61% and 99.09%, respectively.

Keywords:

• Dual tree complex wavelet transform
• feature fusion
• polarimetric SAR
• classification

Data availability statement

No new datasets were generated in this paper. The datasets used for the experiments are publicly available datasets.

Disclosure statement

The author declares that she has no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.