Abstract:High-resolution polarimatric synthetic aperture radar (PolSAR) images can provide more detail information on land-cover types and increase the image complexity at the same time. Traditionally, pixel-based image classification that takes image pixel as a processing unit cannot make full use of various features contained in high-resolution remote sensing images, and thus may not obtain satisfactory results. Hence, object-based image classification (OBIC) methods using image objects as processing units have been introduced into the PolSAR image classification. However, current researches on OBIC methods for PolSAR images usually could not take advantage of multiscale information of image objects, leading to some results that are not as satisfactory as expected. In this article, a multilevel image description consisting of proposed pixel-level spatial and object-level semantic features is developed for OBIC of PolSAR images. At the image pixel level, based on th combination of polarimetric and morphological image descriptions, polarimetric morphological profiles are developed to describe pixel-level spatial features. At the image object level, based on the construction of object adjacent graph, an object-level semantic indicator is proposed, which takes into account the contextual neighborhood of image objects. Finally, the proposed pixel-level spatial and object-level semantic features are integrated and incorporated in an OBIC scheme for the PolSAR image classification. Two fully polarized datasets acquired by ESAR and uninhabited airborne vehicle synthetic aperture radar (UAVSAR), respectively, are adopted to evaluate the effectiveness of the proposed method. The experimental results validate that the comprehensive utilization of both pixel-level and object-level features can effectively improve the OBIC accuracy of PolSAR images.

Polsar Image Classification Based On An Improved Bow Model With Mid-Level Semantic Features

Effects of BOW Model with Affinity Propagation and Spatial Pyramid Matching on Polarimetric SAR Image Classification

Object-Based Classification of PolSAR Images Based on Spatial and Semantic Features

Polsar Image Classification Based On Dbn And Tensor Dimensionality Reduction

Object-Based Classification Method for PolSAR Images with Improved Scattering Powers and Contextual Features

SUPERPIXEL-BASED CLASSIFICATION USING SEMANTIC INFORMATION FOR POLARIMETRIC SAR IMAGERY

PolSAR Image Classification Based on Deep Polarimetric Feature and Contextual Information

POLSAR IMAGE CLASSIFICATION BASED ON POLARIMETRIC OBJECT-BASED MORPHOLOGICAL PROFILES

Robust Classification of PolSAR Images Based on Pinball loss Support Vector Machine

Polarmetric SAR Images Classification Based on Sparse Representation Theory.

The Study of POL-SAR Image Classification Based on SVM

Deep Learning Based Classification Using Semantic Information for Polsar Image

Joint Polarimetric-Adjacent Features Based on LCSR for PolSAR Image Classification

Composite Kernel Method for Polsar Image Classification Based on Polarimetric-Spatial Information

High-Resolution Polsar Image Interpretation Based on Human Images Cognition Mechanism

Fully Polarimetric SAR Image Classification Via Sparse Representation and Polarimetric Features.

Polsar Image Classification Based on Three-Dimensional Wavelet Texture Features and Markov Random Field.

A Graph-Based Semisupervised Deep Learning Model for PolSAR Image Classification

Unsupervised Classification of Polarimetirc SAR Image Via Improved Manifold Regularized Low-Rank Representation with Multiple Features

From Pixel to Superpixel: A Multi-Scale Strategy for Polarimetric Sar Image Classification

Superpixel-Based Classification Using K Distribution And Spatial Context For Polarimetric Sar Images