Abstract
The Canadian RADARSAT constellation mission (RCM) is represented by three synthetic aperture radar (SAR) satellites, each of which includes a compact polarimetry (CP) mode. CP is advantageous because it provides increased backscatter information relative to single and conventional dual-polarized modes and has larger swath widths relative to a quad polarization mode. CP captures single-look complex data which can be used to derive the multilook complex (MLC) coherence matrix, or, equivalently, the Stokes vector data of the backscattered field. The challenge is to develop computer vision algorithms that can be used to effectively segment the scene using this new data source. An unsupervised region-based segmentation approach has been designed and implemented that utilizes the complex Wishart distribution characteristic of the MLC CP data. The segmentation method is based on the iterative region growing with semantics algorithm originally designed for single and dual pol intensity SAR data. The algorithm has been tested using both simulated CP SAR images and a pair of available quad polarization SAR images. The results demonstrate that the CP-IRGS algorithm provides more accurate segmentation images than those using only the RH and RV channel intensity images.
Highlights
S YNTHETIC aperture radar (SAR) offers a remote sensing sensor that is used on orbiting satellites to create digital images of the earth’s surface
Compact polarimetry (CP) is a SAR mode that offers a compromise between traditional dual polarization (DP) and quad polarization (QP) systems
In this paper, we provide a detailed description of the statistical characteristics of the CP data and develop a region-based unsupervised segmentation method for CP data
Summary
S YNTHETIC aperture radar (SAR) offers a remote sensing sensor that is used on orbiting satellites to create digital images of the earth’s surface. There has been a trend towards utilization of CP SAR data for a variety of remote sensing applications such as terrain classification [3], oil spill detection [4], and change detection [5]. The purpose of an image segmentation is essentially to find the optimum label configuration. IRGS is formulated based on the Bayesian theory where the objective is to find a label configuration y∗ that satisfies: y∗ = arg max p(x|y)P (y), (7). The term p(x|y) is called the feature model or data likelihood and is the conditional probability density function of the image data x given the label configuration y. IRGS is a region-based method which uses a region adjacency graph (RAG) [26] and aims to find the optimum label field over a RAG instead of all the image sites separately. A region v ∈ V in the image consists of a set of image sites Sv
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call