Polarimetric features extracted from polarimetric synthetic aperture radar (PolSAR) images contain abundant back-scattering information about objects. Utilizing this information for PolSAR image classification can improve accuracy and enhance object monitoring. In this paper, a deep learning classification method based on polarimetric channel power features for PolSAR is proposed. The distinctive characteristic of this method is that the polarimetric features input into the deep learning network are the power values of polarimetric channels and contain complete polarimetric information. The other two input data schemes are designed to compare the proposed method. The neural network can utilize the extracted polarimetric features to classify images, and the classification accuracy analysis is employed to compare the strengths and weaknesses of the power-based scheme. It is worth mentioning that the polarized characteristics of the data input scheme mentioned in this article have been derived through rigorous mathematical deduction, and each polarimetric feature has a clear physical meaning. By testing different data input schemes on the Gaofen-3 (GF-3) PolSAR image, the experimental results show that the method proposed in this article outperforms existing methods and can improve the accuracy of classification to a certain extent, validating the effectiveness of this method in large-scale area classification.

This study employs the reflection symmetry decomposition (RSD) method to extract polarization scattering features from ground object images, aiming to determine the optimal data input scheme for deep learning networks in polarimetric synthetic aperture radar classification. Eight distinct polarizing feature combinations were designed, and the classification accuracy of various approaches was evaluated using the classic convolutional neural networks (CNNs) AlexNet and VGG16. The findings reveal that the commonly employed six-parameter input scheme, favored by many researchers, lacks the comprehensive utilization of polarization information and warrants attention. Intriguingly, leveraging the complete nine-parameter input scheme based on the polarization coherence matrix results in improved classification accuracy. Furthermore, the input scheme incorporating all 21 parameters from the RSD and polarization coherence matrix notably enhances overall accuracy and the Kappa coefficient compared to the other seven schemes. This comprehensive approach maximizes the utilization of polarization scattering information from ground objects, emerging as the most effective CNN input data scheme in this study. Additionally, the classification performance using the second and third component total power values (P₂ and P₃) from the RSD surpasses the approach utilizing surface scattering power value (P_S) and secondary scattering power value (P_D) from the same decomposition.

Wetlands are the “kidneys” of the earth and are crucial to the ecological environment. In this study, we utilized GF-3 quad-polarimetric synthetic aperture radar (QP) images to classify the ground objects (nearshore water, seawater, spartina alterniflora, tamarix, reed, tidal flat, and suaeda salsa) in the Yellow River Delta through convolutional neural networks (CNNs) based on polarimetric features. In this case, four schemes were proposed based on the extracted polarimetric features from the polarization coherency matrix and reflection symmetry decomposition (RSD). Through the well-known CNNs: AlexNet and VGG16 as backbone networks to classify GF-3 QP images. After testing and analysis, 21 total polarimetric features from RSD and the polarization coherency matrix for QP image classification contributed to the highest overall accuracy (OA) of 96.54% and 94.93% on AlexNet and VGG16, respectively. The performance of the polarization coherency matrix and polarimetric power features was similar but better than just using three main diagonals of the polarization coherency matrix. We also conducted noise test experiments. The results indicated that OAs and kappa coefficients decreased in varying degrees after we added 1 to 3 channels of Gaussian random noise, which proved that the polarimetric features are helpful for classification. Thus, higher OAs and kappa coefficients can be acquired when more informative polarimetric features are input CNNs. In addition, the performance of RSD was slightly better than obtained using the polarimetric coherence matrix. Therefore, RSD can help improve the accuracy of polarimetric SAR image classification of wetland objects using CNNs.

In synthetic aperture radar (SAR) image sea ice classification, the polarization decomposition techniques are used to enhance classification accuracy. However, traditional methods, such as Freeman–Durden (FD) and H/A/α decomposition, struggle to accurately characterize complex scattering mechanisms, limiting their ability to differentiate between various sea ice types. This paper proposes using the Modified Reflection Symmetry Decomposition (MRSD) method to extract polarization features from Gaofen-3 (GF-3) satellite fully polarimetric SAR data for sea ice classification tests. The study data included three types of sea surface: open water (OW), young ice (YI), and first-year ice (FYI). In this research, backscattering coefficients were combined with FD, H/A/α, and MRSD polarization features to create eight feature combinations for comparative analysis. Three machine learning algorithms, Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Machines (SVM), were also used for the comparative analysis. The results show that MRSD polarization features significantly improve model performance, particularly distinguishing among sea ice categories. Compared to using only the backscatter coefficient, MRSD polarization features increased model classification accuracy by approximately 4% to 13%, outperforming FD and H/A/α polarization features. The XGBoost model trained with MRSD polarization features achieves excellent classification results, with classification accuracies of 0.9630, 0.9126, and 0.9451 for OW, YI, and FYI. Additionally, the model achieved a Kappa coefficient of 0.9105 and an F1-score of 0.9403. Feature importance and SHapley Additive exPlanations (SHAP) analysis further demonstrate the physical significance of the MRSD polarization features and their role in model decision-making, suggesting that the scattered component power plays a crucial role in the model’s classification decision. Compared to traditional decomposition methods, MRSD provides a more detailed characterization of scattering mechanisms, offering a comprehensive understanding of the physical properties of sea ice. This paper systematically demonstrates the superior effectiveness of MRSD polarization features for sea ice classification, presenting a new scheme for more accurate classification.

For a multi-look polarimetric synthetic aperture radar (POLSAR) image, each pixel corresponds to a polarimetric coherency matrix. Model-based incoherent polarimetric decomposition is a technique which is widely used to analyze multi-look POLSAR data. Traditional model-based incoherent polarimetric decomposition algorithms have some inherent drawbacks such as negative power components, polarimetric information loss, and non-model-based decomposition results. This study tries to completely interpret a polarimetric coherency matrix by the incoherent sum of four scattering mechanisms. Therefore, the proposed algorithm can be regarded as a new type of model-based incoherent polarimetric decomposition. All the four scattering models are firstly derived with polarimetric symmetry. The four scattering models correspond to surface scattering, double-bounce scattering, volume scattering and helix scattering, respectively. Then a new four-component model-based incoherent decomposition algorithm is found. After extracting the helix scattering component and the maximum possible volume scattering component, the remaining coherency matrix is decomposed into two components with an orientation angle difference of 45°. With the new algorithm, most pixels of a real POLSAR image can be completely decomposed into four components which are exactly consistent with helix scattering, volume scattering, surface scattering, and double-bounce scattering, respectively. Moreover, the proposed decomposition algorithm fully utilizes the polarimetric information, and all scattering component powers are nonnegative. Experiments with E-SAR, RADARSAT-2, and GF-3 data are presented to illustrate the effectiveness of analyzing the scattering mechanisms of real terrain targets with the proposed decomposition algorithm. The proposed decomposition algorithm is also compared with classic four-component model-based incoherent polarimetric decomposition algorithms.

Freeman−Durden Decomposition (FDD) and Non-Negative Eigenvalue Decomposition (NNED) are among the most widely used incoherent polarimetric decomposition algorithms for analyzing fully Polarimetric Synthetic Aperture Radar (PolSAR) data, particularly FDD. However, with advancements in model-based incoherent polarimetric decomposition techniques, their original algorithms have certain aspects that can be modified to enhance their decomposition performance. These aspects include: FDD occasionally yielding negative power values and typically overestimating the power of the volume scattering component; the scattering mechanism of the remainder matrix in NNED being further interpretable; and the potential for improving its decomposition performance through specific modifications. Therefore, two improved incoherent polarimetric decomposition algorithms, Modified Freeman−Durden Decomposition (MFDD) and Deorientation Non-Negative Eigenvalue Decomposition (DNNED), are proposed in this study. For MFDD, deorientation is applied at the outset, and two additional steps are introduced to eliminate negative power values in the decomposition results. The DNNED algorithm also employs deorientation and enhances the interpretation of the scattering mechanism of the remainder matrix. DNNED identifies that the remainder matrix corresponds to a dihedral with a 45-degree orientation angle, thus classifying its power as double-bounce scattering. Decomposition performance tests have been conducted using two actual PolSAR images derived from E-SAR of Germany and GF-3 of China. Experimental results demonstrate that the performance of MFDD is superior to that of FDD, and the performance of DNNED is the best among the four aforementioned decomposition algorithms.