PolInSAR Coherence and Entropy‐Based Hybrid Decomposition Model

Abstract

Target characterization is an essential aspect of polarimetric decomposition. This technique is capable of categorizing polarimetric signatures for different types of targets based on the scattering mechanisms they follow, enabling straightforward physical interpretation of the targets. The geometric anomalies associated with human-made targets escalate the degree of randomness in the scattering process, which causes scattering ambiguity for such targets. The second-order model descriptors do not relate to the actual physical structure and yield predominant volume scattering power. Such urban targets are decomposed as natural targets leading to irrelevant decomposition results. The methods developed to curb the problem are unable to maintain the consistency in the decomposition modeling as they underestimate volume scattering powers for natural landcover. A hybrid decomposition model is proposed herein to solve the problem of predominant volume scattering observed from urban targets by preserving volume scattering powers for natural targets. The model uses eigenvalue-based decomposition parameters and polarimetric interferometric synthetic aperture radar (PolInSAR) coherence to decompose ambiguous targets. The proposed model has been tested on NISAR UAVSAR PolInSAR data acquired over the Greenville region, MS, USA. The proposed model has increased the double-bounce scattering from the urban targets and enhanced the volume scattering from natural landcover as well. By comparing the results with existing decomposition models, it is observed that the proposed model gives a more robust representation of the landcover than the compared decomposition models. Plain Language Summary The SAR sensor illuminates the surface with polarized microwaves and receives the interacted backscattered wave. The backscattered wave is transformed into information, which is represented by a scattering matrix. Due to different geophysical properties of the targets, there is a change in the backscattered wave. Polarimetric SAR decomposition is a technique that can relate the change in backscatter to the physical structure of the target, enabling simple and robust classification of the landcover. The techniques decompose or disintegrate the scattering matrix into scattering mechanisms, and each scattering mechanism represents a broad category of different landcover. Owning to the random nature of real-world observations, the backscatter for two different categories of landcover can be the same, such as a building and a tree. Therefore, for such cases, the polarimetric decomposition yields inaccurate classification results, and from the polarimetric point of view, this problem is referred to as overestimation of volume scattering. The reason for this is the assumptions and the design of the model. Various modifications have been proposed to solve the problem of overestimation. We find that direct modification may yield inconsistent decomposition results, which can be critical for different utilities of polarimetric decompositions. We propose...

Estimation of aboveground biomass from PolSAR and PolInSAR using regression-based modelling techniques

“…PolSAR is sensitive to different parameters of objects such as shape, orientation, dielectric properties, as well as permittivity, material constituents of the target and also the ensemble average entropy, which can be used for measuring forest biophysical parameters (Rencz and Ryerson 1999;Papathanassiou and Cloude 2001;Krieger et al 2005). The PolSAR scattering mechanisms contributed from each SAR resolution cell can be retrieved separately after decomposition (Shafai and Kumar 2020;ME and Kumar 2021). The scattering mechanisms are odd-bounce, volume, double-bounce, helix and wire scatterings (Zhang et al 2008).…”

Semantic segmentation of PolSAR image data using advanced deep learning model

“…To retrieve all the scattering information contributed by different objects in a small area, fully polarimetric quad-pol data are used because retrieval of all the scattering elements within a resolution cell is mathematically and practically rigorous with single or dual polarised SAR data 62 , 63 . The remote sensing community has widely used the fully polarimetric quad-pol SAR data to implement model-based decomposition approaches in the retrieval of scattering parameters 64 – 66 . The accuracy in polarimetric SAR model-based scattering retrieval is affected by the polarimetric distortion of the SAR data 67 , 68 .…”

Dielectric characterization and polarimetric analysis of lunar north polar crater Hermite-A using Chandrayaan-1 Mini-SAR, Lunar Reconnaissance Orbiter (LRO) Mini-RF, and Chandrayaan-2 DFSAR data

“…Barnes decomposition is based on the 'wave dichotomy' and H-A-Alpha decomposition is based on eigenvalue and eigenvector . The use of these two decompositions in this study was to minimize the overestimation of particular scattering (Shafai & Kumar, 2020;Bhanu Prakash & Kumar, 2021a, 2021bRamya & Kumar, 2021) in a target area which is easily seen in the modelbased decomposition techniques.…”