Retrieval-based multimodal document QA aims to identify and integrate relevant information from visually rich documents with complex multimodal structures. While retrieval-augmented generation (RAG) has shown strong performance in text-based QA, its extensions to multimodal documents remain underexplored and face significant limitations. Specifically, current approaches rely on query-agnostic document representations that overlook salient content and use static top-k evidence selection, which fails to adapt to the uncertain distribution of relevant information. To address these limitations, we propose the Multimodal Adaptive Retrieval-Augmented (MARA) framework, which introduces query-adaptive mechanisms to both retrieval and generation. MARA consists of two components: a Query-Aligned Region Encoder that builds multi-level document representations and reweights them based on query relevance to improve retrieval precision; and a Self-Reflective Evidence Controller that monitors evidence sufficiency during generation and adaptively incorporates content from lower-ranked sources using a sliding-window strategy. Experiments on six multimodal QA benchmarks demonstrate that MARA consistently improves retrieval relevance and answer quality over existing SOTA method.

For synthetic aperture radar (SAR), multipolarization and multifrequency modes greatly enrich the acquired earth resource information and have been widely applied in remote-sensing fields. In this article, we compare the classification capabilities of multipolarization and dual-frequency SAR. To meet the objective of selecting consistent and complete polarimetric information, a unified classification framework is proposed. In the framework, covariance matrices are used directly as inputs instead of polarimetric indicators. Additionally, the Wishart mixture model (WMM) is utilized to characterize the statistical distribution of polarimetric SAR data. Besides, the data log-likelihood function is utilized to mitigate the influence of the initial values on the expectation-maximization (EM) algorithm. Then, among the combinations of four sample-to-subclass distances and two schemes for obtaining sample-to-class distances, the one with the best classification performance is selected as the default for this framework. In the experiments, the classification capabilities of full polarization (FP), compact polarization (CP), and dual-polarization (DP) modes are first compared through the proposed classification framework. Then, we compare the classification capabilities of dual-frequency SAR in FP, CP, and DP modes. For polarimetric SAR (PolSAR) system design, it is necessary to strike a balance between demand indexes (classification performance, coverage width, and so on) and cost (such as budget, weight, and so on). The comparison results provide a reference for the optimization of polarization modes and frequency bands of the existing multipolarization and dual-frequency SAR payloads and the design of future SAR systems.

In this paper, we investigate the potential of polarimetric synthetic aperture radar (SAR) in identifying flat areas using fractal dimension and polarimetric scattering similarity. A two-step method is proposed, including rough selection and fine selection. First, rough selection is performed by calculating the fractal dimension of the radar backscattered total power image. Then for each candidate region, the fine selection is conducted using polarimetric scattering similarity parameters. Furthermore, the effectiveness of the method is verified by GF-3 quad-polarimetric SAR data and SRTM1 DEM data in desert areas of China. Results show that for the final selected flat area (320 × 320 m), the maximum elevation deviation is 3.39 m and the elevation standard deviation is 0.72 m. Therefore, without depending on additional DEM data, the proposed method can effectively achieve flat areas identification, which can be helpful for the future application of polarimetric SAR data in the Moon.

In this letter, a four-component decomposition method of polarimetric synthetic aperture radar interferometry (PolInSAR) using refined volume scattering models is proposed. In the proposed algorithm, the volume scattering models under the assumption of reflection symmetry are refined by employing the orientation angles. In addition, the polarimetric interferometric similarity parameter (PISP) parameter is introduced to modify the refined volume scattering models. Airborne L-band PolInSAR data are used to evaluate the performance of the method. The experimental results demonstrate that the proposed method can effectively overcome the overestimation of the volume scattering and characterize the scattering mechanisms of various terrain types.

In this letter, a hierarchical extended multicomponent decomposition method (ExMCSM) of polarimetric synthetic aperture radar interferometry (PolInSAR) is proposed. The target of this method is to overcome the overestimation of volume scattering (OVS) and to mitigate the mixed ambiguity of the scattering mechanism. In the proposed method, prior to the decomposition, orientation angle compensation (OAC), and helix angle compensation (HAC) are applied to the coherency matrix to reduce the complexity of the operation. The polarimetric interferometric similarity parameter (PISP) is used to refine the volume scattering models, and the optimal coherence magnitude (γopt3) is used as a criterion for the adaptive selection of these volume scattering models. The efficiency of the proposed method is demonstrated using two different test sites. Experimental results demonstrate that the proposed method can effectively represent the scattering characteristics of the ambiguous regions.

Due to the variability of urban structures, buildings with significant cross-polarized scattering may be misclassified as forests. Therefore, urban area extraction is still a challenging problem. In this paper, a new urban extraction method using optimized roll-invariant features and selected hidden polarimetric features in the rotation domain is proposed. The optimal coherency ratio along radar line of sight is calculated by traversing the rotation angle within permissible values, and hidden features are selected based on the similarity processing and principal component analysis (PCA) algorithm in the rotation domain. The fusion of correlated probabilities (FCP) algorithm is applied to commendably enhance the classification accuracy. Spaceborne Gaofen-3 full PolSAR data is used to validate the performance of the proposed method. Experimental results demonstrate that the proposed method is capable of extracting urban areas from natural areas.

In this paper, a hierarchical extended adaptive general four-component decomposition method (AG4U) is proposed to solve the overestimation problem of volume scattering (OVS). In the proposed method, the transformed coherency matrix and branch conditions are determined by choosing one of the two unitary transformation matrices, and the refined volume scattering model is transformed adaptively in order to adapt to various terrain types. Spaceborne Gaofen-3 full PolSAR data is used to verify the effectiveness of the proposed decomposition method. Comparison experiments show that the proposed method can effectively represent the scattering characteristics of the ambiguity regions, and the oriented building areas can be well discriminated as dihedral or odd bounce structures.

A fully polarimetric response of landslide areas at X-band was studied by a Chinese high-resolution airborne synthetic aperture radar system. Polarimetric decompositions, including the Yamaguchi four-component decomposition and the Cloude decomposition, are used to analyze the scattering mechanisms of several typical landslides caused by the 2008 Wenchuan Earthquake in southwestern China. The experimental results indicate that areas affected by large-scale landslides show complicated scattering mechanisms at X-band, which are a mixture of surface, double bounce, and volume scattering. Simple classification results based on supervised Wishart classifier and polarimetric scattering similarity parameters are also presented, which can distinguish landslide areas from others, such as forest and water, very well. However, it does not perform well for urban areas. Additional information, such as prelandslide imagery, is needed to distinguish landslide areas from urban areas or bare soil. From these results, we can conclude that landslide mapping using fully polarimetric data has great potential for rapid response and management of landslide disasters.

Nonnegative eigenvalue decomposition (NNED), which insists and guarantees that each decomposed scattering component corresponds to a physically realizable scatterer, is powerful for polarimetric synthetic aperture radar (SAR) images analysis. Previous NNED is mainly illustrated under the reflection symmetric condition. In this paper, the coherency matrix approach is derived to implement the NNED for the nonreflection symmetry scattering case. We explicitly show the diversifications of the decomposition results between NNED with and without reflection symmetry assumptions, and quantitatively analyze the differences between them using the E-SAR polarimetric data acquired over the Oberpfaffenhofen area in Germany.

Polarimetric calibration is aimed at intercalibrating the polarization channels of the radar system. A better characterization of the target and of the backscattering mechanisms can be obtained by polarimetric calibration. The study described in this paper is an attempt at deriving the requirements on the polarimetric calibration for a commonly used classification method based on the scattering type alpha. A model representing the effects of the residual distortion of the polarimetric data after calibration is present. This distortion model is then applied to calculate alpha angle, through which the polarimetric calibration requirements for identifying seven scatterers can be obtained. We will show that, the phase of parameter f, representing the phase difference of the mismatch between the like-polarized channels, affects the classification results most significantly.