Synthetic aperture radar (SAR) images obtained from multi-sensor systems usually exhibit significant shift in data distribution, known as the domain shift. It is challenging to utilize the relevant knowledge of multi-sensor datasets for SAR image cross-domain classification with general supervised learning methods. While domain adaptation (DA) methods can alleviate the domain shift by aligning distributions, they are limited in handling the scale/resolution variations observed in multi-sensor SAR images. These methods mainly focus on feature representations at a single scale for distribution alignment, which may fail to fully align the distributions across different scales. To solve this problem, we propose a new multi-scale alignment domain adaptation network (MSADAN) for SAR ship cross-domain classification. MSADAN explicitly considers the scale factor, enabling us to overcome the weak generalization observed in existing DA methods when dealing with SAR images exhibiting significant scale/resolution variations. Specifically, we develop a scale-aware feature extractor to effectively capture the multi-scale information present in the datasets, facilitating comprehensive representation learning. Furthermore, we propose a multi-level bilinear fusion (MLBF)-based adversarial learning strategy to overcome the limitations of single-scale feature extraction for distribution alignment, aiming to enhance the generalization ability of the model across domains. In addition, a customized class contrastive loss is designed to improve the inter-class separability and intra-class compactness by penalizing cross-class confusion. Experimental results on datasets demonstrate the superiority of MSADAN in SAR ship cross-domain classification.Note to Practitioners—Ship classification using Synthetic aperture radar (SAR) images proves to be a promising strategy in modern maritime monitoring systems. The primary motivation of this paper is to develop a cross-domain classification system tailored for SAR ship target. In real-world scenarios, SAR data often presents significant resolution variations due to the different imaging modes and conditions across multiple sources. Existing DA methods fail to effectively address this unique challenge in the remote sensing field, resulting in poor cross-domain classification performance. The proposed method considers the scale or resolution variations of the targets during feature learning. Furthermore, explicitly incorporating resolution into the distribution alignment enhances the generalization ability of the model to target tasks when dealing with the significant scale or resolution variations across datasets. Experimental results confirm the practicality and robustness of our SAR ship cross-domain classification method in real scenarios. This is of significant importance in promoting the application of machine learning approaches in practical scenarios. In the future, we plan to integrate the physical scattering information of SAR images for the interpretable cross-domain classification for SAR targets.

Cross-scene classification aims to transfer knowledge acquired from a label-rich source domain to an unlabeled target domain with a distribution shift. With the large amount of available remote sensing data from diverse satellite platforms, it prompts us to utilize the knowledge from extensive datasets to address target tasks in small-scale domains, known as partial domain adaptation (PDA). However, the PDA setting poses significant challenges for remote sensing scene images. Existing methods often fail to sufficiently explore both task-specific and transferable knowledge across domains based on the representations of entire samples, potentially resulting in the amplification of negative transfer brought by irrelevant knowledge. To address this, we propose a new selective alignment transformer (SAT) designed to distinguish transferable and untransferable knowledge across domains for cross-scene classification in RSIs under the PDA scenario. Specifically, a new bi-level reweighting strategy that incorporates transferability-aware patch selection and class-wise reweighting is developed to emphasize the transferable image patches and classes. Based on the aforementioned reweighting strategy, we further introduce a patch-weighted maximum mean discrepancy (PMMD) loss, which selectively aligns the distributions from the patch-level perspective, facilitating the learning of transferable domain-invariant representations. The experimental results of SAT demonstrate its effectiveness and superiority in addressing this practical domain adaptation (DA) task, outperforming state-of-the-art methods in PDA tasks on four datasets.

Height is an important indicator of rice growth condition and phenological period. The Polarimetric SAR Interferometry (PolInSAR) has proven a valuable technique for providing structural metrics of vegetation. Anechoic chamber experiments, allowing a high flexibility in the configuration of operating parameters, such as frequency and baseline, is the most convenient and feasible way to study the potential of PolInSAR in rice height retrieval. In this study, the first PolInSAR measurements of a recently built anechoic chamber, The Laboratory of Target Microwave Properties (LAMP), were presented and analyzed versus frequency and baseline, and rice height was retrieved using the measurement data. The results confirm that LAMP has the ability of PolInSAR, and the height of rice in milk stage is retrieved more precisely in the frequency range of 2.5 to 4.7GHz, when the baseline is 0.5°. The results are expected to contribute to studying and to defining the operation modes for future SAR missions.

In synthetic aperture radar (SAR) automatic target recognition (ATR), there are mainly two types of methods: physics-driven model and data-driven network. The physics-driven model can exploit electromagnetic theory to obtain physical properties, while the data-driven network will extract deep discriminant feature of targets. These two types of features represent the target characteristics in scattering domain and image domain, respectively. However, the representation discrepancy caused by the different modalities between them hinders the further comprehensive utilization and fusion of both features. In order to take full advantage of physical knowledge and deep discriminant feature for SAR ATR, we propose a new feature fusion learning framework SDF-Net to combine scattering and deep image features. In this work, we treat the attributed scattering centers (ASC) as set-data instead of multiple individual points, which can well mine the topological interaction among scatterers. Then multi-region multi-scale sub-sets are constructed at both component and target levels. To be specific, the most significant scattering intensity and overall representation in these sub-sets are exploited successively to learn permutation-invariant scattering features according to a set-oriented deep network. The scattering representations can provide mid-level semantic and structural features that are subsequently fused with the complementary deep image features to yield an end-to-end high-level feature learning framework, which helps enhance the generalization ability of networks especially under complex observation conditions. Extensive experiments on Moving and Stationary Target Acquisition and Recognition database verify the effectiveness and robustness of the SDF-Net compared against both typical SAR ATR networks and ASC-based models.

Radio signals emitted by various sources, such as ground radars and broadcast/communication devices, can unintentionally cause radio frequency interference (RFI) to spaceborne synthetic aperture radar (SAR), degrading SAR image qualities to various degrees. Most existing methods tackle this problem by applying specially designed preprocessing steps to RFI-polluted level-0 SAR data before SAR focusing. However, such preprocessing is not widely used in spaceborne SAR, as there exist radiometric artifacts due to various RFI sources in the level-1 single-look complex (SLC) image products in many spaceborne SAR data, e.g., Sentinel-1 open data archives. To address this problem, in this article, we first propose a generic subspace model for characterizing a variety of RFI types, which reveals a low-dimensional structure of RFI subspace. Based on the proposed model, we next design a block subspace filter (BSF) for removing RFI artifacts in SLC SAR images directly. Experiments with ERS-2, ENVISAT/ASAR, Sentinel-1, and Gaofen-3 data are presented, and quantitative assessments based on numerical simulations are provided, which demonstrates the promising performance and application potentials of the proposed method. BSF is simple yet efficient and does not require performing preprocessing on level-0 raw data, which is helpful for users to obtain clean SAR images. MATLAB/Octave code implementation of BSF is available athttps://github.com/huizhangyang/BSF.

Information of crop phenology is essential for evaluating crop productivity and crop management. Synthetic Aperture Radar (SAR), with the advantage of all-weather, day-night imaging and clouds penetrability, is an effective way for rice growth monitoring. In this study, we developed a method for remotely determining phenological stages of paddy rice with sixteen polarimetric SAR images. The method consists of three procedures: (I) Classified the transplanted rice (T-R) and the direct-sown rice (D-R) field; (II) Sensitivity analysis of polarimetric parameters versus rice phenology; (III) Reconstructing the time-series polarimetric parameters profiles of rice by time-frequency analysis; (IV) Specifying the phenological stages by detecting the maximum point, minimal point and inflection point from the smoothed polarimetric parameters time profile. Three keys phenological periods of rice were detected with the the accuracy of about 84%.

This study evaluated the potential of a compact polarimetric (CP) mode SAR concept for the retrieval of winter wheat LAI values. First, the CP SAR data used in this study were simulated from full polarimetric SAR data. A total of 19 CP parameters were extracted from CP matrices and from CP decomposition results. Second, a hybrid inversion model based on nonlinear genetic partial least square (GA-PLS) algorithm was built to determine the relationships between CP SAR parameters and LAI values. Finally, the mapping and estimation of winter wheat LAI using the CP SAR parameters by GA-PLS algorithm were implemented. Our study indicates that by using a CP SAR dataset, LAI retrieval can yield an accuracy of R2=0.70 and RMSE = 0.40 m2/m2. Our study has significant implications for field crop condition monitoring and growth stage estimation, providing relevant data for agronomic practices and precision agriculture.

The objective of this paper is to evaluate the potential of Gaofen (GF)-3 quad-polarized Synthetic Aperture Radar (SAR) for coastal wetland observation. The north coastal wetland of Zhejiang is selected as the test case, which is operated by on-site measure as reference results, matched up with three scenes of GF-3 Quad Polarized Strip I (QPSI) mode SAR imagery. Based on the collected datasets, three well-known methods of Pauli, Freeman and H/A/Alpha decomposition are performed to data processing, and the experimental results reveal that reliable observation capability of GF-3 quad-polarized SAR is verified. The promising preliminary results indicate that GF-3 is encouraging for operational implementation, especially for coastal wetland observation.

The use of a microwave scattering model is fundamental to rice monitoring by synthetic aperture radar (SAR), and an ear scattering model is necessary to improve the accuracy of the microwave scattering model of the rice field. A novel rice ear scattering model, the multisphere ear scattering model, was developed by incorporating the microstructure of rice ear panicles, including the ear grain parameters in particular. A virtual ear model was used to simulate the ear morphology, and a multisphere scattering model was used to simulate the ear scattering. The multisphere ear scattering model has the potential to retrieve grain parameters from SAR imagery. The new model was also compared with a traditional two-cylinder ear model; the large discrepancy in the simulation results shows the necessity of carrying out a verification experiment.

Scattering from rice leaves contributes substantially to total vegetation canopy backscattering and detailed knowledge about it is necessary for developing a microwave scattering model. A parabolic curve is generally adopted to simulate the leaf shape but this is rarely incorporated into the calculation of the scattering. In this paper, two specific models, one based on physical optics (PO) approximation and the other on the discrete dipole approximation (DDA), are presented to involve the parabolic leaf curvature effects. Three typical leaves were chosen from 1433 parabolic leaves obtained during ground measurements. The PO and DDA models were used to calculate the leaf scattering. The generalized Rayleigh-Gans (GRG) approximation was also included in the simulation. The method of moments, a computational electromagnetic method, was utilized to evaluate the accuracy of each model. Validation of the models was conducted at incidence angles ranging from 10° to 60°, incidence azimuthal angles ranging from 0° to 360°, and incidence frequencies of 1.2 GHz (L-band), 5.4 GHz (C-band), and 9.65 GHz (X-band). Among the GRG approximation, the DDA model and the PO model, the latter gave the best accuracy ->65% in the cases tested, while the GRG model was the least accurate. The high accuracy of the PO model was maintained at both the low and high frequency bands. The PO model, therefore, has great potential for use to interpret radar measurements from rice fields and other types of vegetation canopy.

Rice fields occupy a land area of more than 100 million ha in Asia, feed about 3.5 billion people worldwide. There is a significant interest in the information provided by remote sensing about rice fields in a timely and efficient manner. Information on phenological development is a fundamental key to rice monitoring because it describes the actual state of the rice plants and their relation with the pedoclimatic conditions. As part of an ongoing effort to explore the use of polarimetric SAR data in rice monitoring, this study proposed an automated method of rice phenology retrieval using a feature optimization strategy integrated support vector machine (SVM) and sequential forward selection (SFS). The optimal polarimetric variables for each phenological stage were acquired, based on which eight rice phenological stages were retrieved automatically. The accuracies were higher than 90% except for the dough stage and the transition period between two stages.

Rice leaf scattering contributes substantially to total vegetation canopy backscattering. A modified Physical Optics (PO) approximation model and a specified Discrete Dipole Approximation (DDA) model have been developed that involves parabolic leaf curvature effects. Three typical leaves are chosen among 1433 pieces of parabolic leaves sampled in a ground measurement. PO, DDA, Generalized Rayleigh Gans (GRG) approximation, were used to calculate the leaf scattering, respectively. The method of moments (MOM), a computational electromagnetic method, is utilized to evaluate each models' accuracy. Validation work is conducted under the incidence angles 10° ~ 60°, incidence azimuthal angle 0° ~ 360°, incident frequencies L Band (1.2 GHz), C Band (5.4 GHz) and X Band (9.65 GHz). Compared with DDA and GRG, the modified PO model presents the maximum accuracy in >84% testing cases, and retains the high accuracy in both the low and high frequency band.

This paper proposed a modified Water Cloud Model (MWCM) for rice variable estimation during the whole growth season with eight RADARSAT-2 quad-pol SAR images. The improvements achieved with the MWCM include considering the heterogeneity of water content of the rice canopy in different directions and different phenologies, and applying the scattering components from an improved polarimetric decomposition in the model instead of the backscattering coefficients. With the MWCM, four rice variables were estimated through the genetic algorithm, including leaf area index (LAI), rice height (h), volumetric water content of total canopy (mv) and ear biomass (De). The validation was conducted using the field data with the average R2of each variable above 0.8. The median relative error (MRE) of the rice variables ranged from 9% to 15% in most phenological stages. The results demonstrated that the MWCM works well for the estimation of rice biophysical parameters with polarimetric SAR data, and it is significant to consider the heterogeneity of water content of the rice canopy in the horizontal direction for estimation of rice variables during the whole rice growth season.

The ear layer, which is a component of rice, is crucial to rice monitoring and yield estimation. By adding in the ear layer, we have extended the original coherent microwave scattering model, which is based on both the first-order solution of the radiation transfer equation and Monte Carlo numerical simulation methods, to full stage of rice. The detailed scene generation and the geometrical description of rice elements are presented. The propagation path of scattering in rice canopy is reduced. Two approximation methods are used to fit the curving ear by straight cylinders. Ground truth measurements of rice fields in heading stage, including the curvature of ear, were acquired extensively at Jinhu, Jiangsu in eastern China. Measured parameters are used in the new extended model to calculate the C-band backscattering coefficients of rice field. The simulation results are used for comparison with the backscattering coefficients extracted from RADARSAT-2 images to test the validity of the coherent scattering model with the mean absolute error reaching <; 3.3 dB in the copolarization mode. Theoretical analysis reveals that the ear morphology can largely affect the backscattering behavior of the rice field.

China is the largest rice producer in the world. Guizhou province is an important rice growing area in the southwest of China. However, rice monitoring with remote sensing data has great difficulties in this region due to its perennial cloud-coverage weather and undulating terrain. With the emergence of polarimetric SAR data and state of art methods for polarization information extraction, rice monitoring in this region is more promising. In this study, multi-temporal RADARSAT-2 polarimetric SAR data set was acquired in Guizhou, China. The Freeman-Durden, Cloude-Pottier and the Touzi decompositions were used for classification and rice scattering mechanism analysis.

Two layer water cloud model (WCM2), which is a refined version of the conventional water cloud model (WCM), considering the vertical inhomogeneity in the vegetation layer. The vertical inhomogeneity of the vegetation layer is described by the distribution of water content per unit volume. A piecewise linear function is used to describe the distribution of water content per unit volume. We analyze the variation tendency of vegetation scattering under different vertical inhomogeneity conditions by means of WCM2 and validate it using a physically-based model developed at Tor Vergata University, Rome, Italy. The result shows that the vertical inhomogeneity affects vegetation scattering significantly. Comparison between model predications and field measurements on radar backscattering coefficients for soybean shows WCM2 has the potential to get a better prediction result.

This paper investigates the S-band backscattering variation of wheat in its four growing seasons, with the purpose of providing a reliable method for wheat identification and monitoring using satellite S-band synthetic aperture radar (SAR) data. The study is based on the tower-based scatteromerer experiment conducted on the wheat fields at the remote sensing test site of Institute of remote sensing applications Chinese academy of sciences. During the entire growing season, we carried out four experiments and got a great amount of backscatter data including the HH and VV polarization. Synchronously, a wide range of plant parameters, such as biomass, soil moisture and canopy height were measured. This paper describes these experiments and analyzes wheat temporal backscattering variation at S-band with corresponding ground parameters.

In the southwest of China, Synthetic aperture radar (SAR) is anticipated to provide an important tool for forestry inventory because of its all weather capabilities. In this paper, Zhazuo area in Guizhou Province of southwest China, with typical Karst landform, was selected as the test site, and six RADARSAT-2 polarimetric images were used for experiments. Methods for forest mapping based on polarimetric decomposition and multi-temporal polarimetric SAR data fusion were proposed. Experiments showed that polarimetric signatures of forest were significantly different with other targets, and fusion of multi-temporal RADARSAT-2 images can effectively improve image quality and enhance forest and deforestation information.

As the high resolution radar satellite's has been successfully launched, its ability of the typical target's recognition monitor enhances a lot. This article mainly does analysis based on the RADARSAT2 quad-polarimetric data, compared scattering properties of typical target feature with two temporal full polarized data, and used the measured data to drive MIMICS (Michigan Microwave Canopy Scattering model) model and carry on the multi-wave band full polarization backscattering simulation about the farmland and forest. Then we carried on the comparison between X, C band SAR image gain's actual scattering value and simulation value in order to infer the typical target feature' scattering properties rule of S band. With anticipation of HJ-1-C satellite soon launched by China and then we can carrry on the comparison with the fact. Along with the full polarized data's development, the typical target feature's scattering properties analysis will be more perfect.

In the south of China, synthetic aperture radar (SAR) provides a powerful tool for forestry inventory because of its all-weather and all-day capabilities. Nevertheless previous single or dual polarization SAR data cannot meet the requirements of forest type classification. Polarimetric SAR data contained more information of targets and in this paper we investigated the capability of polarimetric Radarsat 2 data for forest type discrimination. Taking Zhazuo forest farm of Guizhou Province as study area, an 8-temporal field experiment was designed and used for polarimetric backscattering signatures analysis based on MIMICS model. Then two-temporal polarimetric Radarsat 2 data was analyzed to extract polarimetric variables for forest species discrimination, and then polarimetric decomposition and classification were carried out. Experiments prove that forest type can be discriminated using polarimetric Radarsat 2 data, but it is not very effective for forest species identification mainly due to the spatial resolution limitation. Polarimetric SAR data with higher resolution and more complicated classification methods are needed in the future.

Rice is a major food supply in the southeast of China, which is very important for this region's rapid and sustainable development. Synthetic Aperture Radar (SAR) provides a powerful tool for rice monitoring in these regions because of its all-weather, day-and-night imaging and canopy penetration capabilities. HJ-1 small satellite constellation of China has been designed for environment and disaster monitoring, and HJ-1-C satellite has a SAR system working in S-band with incidence varying from 31° to 40°, VV polarization. Scattering model is helpful to better understand the temporal behavior of rice backscatter in S-band before HJ-1 SAR satellite is launched. In this paper, Zhaoqing test site in Guangdong province was selected as the test site, and 9-temporal field measurements acquired during the rice growing period in 1997 were used for analysis. Then rice backscattering and seasonal variation in S-band and VV polarization were simulated and analyzed based on radiative transfer model and ground measurements.