High-resolution soil moisture (SM) products are of broad scientific interest and practical application value. The flexibility of estimating soil moisture from Sentinel-1 data has been widely recognized. However, the backscatter coefficient of synthetic aperture radar (SAR) data is heavily influenced by noise during radiation transmission as well as vegetation cover. This directly leads to the inability to accurately achieve high-resolution SM estimation using a single SAR data. In this paper, we proposed a deep learning model that fuses multisource data that estimates the SM in the depth of ~5 cm. The inputs to the model integrated remote sensing data and soil information data. The model was trained and validated on data from in-situ sensors of the international soil moisture network (ISMN). The proposed method achieved a coefficient of determination (R2) of 0.732 and a root mean square error (RMSE) of 0.069. In addition, the validation experiments in Anhui Province, China also demonstrated the effectiveness and robustness of the proposed method with a R2of 0.85.

Reliable and accurate rice mapping using synthetic aperture radar (SAR) in cloudy and rainy areas is essential for achieving the United Nations Sustainable Development Goal 2 of 2030. An interpretable deep learning SAR rice area mapping method is proposed in this letter to suppress the interference of wetlands and other land covers to multi-temporal SAR rice area mapping and improve the accuracy and confidence of the "black box" deep learning model results. Combining the temporal backscattering coefficients and interferometric coherence, three interpretable temporal features are extracted to effectively distinguish rice. Then, the explainable feature-aware network (XFANet), which can provide the learned importance weights of the normalization methods as self-interpretation, is constructed, and the pixel-wise gradient-weighted class activation mapping (PGCAM) post-hoc interpretation method is introduced to interpret the feature variation within the model. The experimental results in the Kampng Chhang and Kampng Chham provinces of Cambodia show that the proposed three interpretable features well suppressed the wetland disturbance to rice. With high interpretability, the overall accuracy of XFANet reaches 93.43%.

As one of the most important grains worldwide, paddy rice plays a significant role in not only regional food supply but also international food trading. In past decades, Synthetic Aperture Radar (SAR) has been proved to be an effective tool in paddy rice monitoring. However, in Southeast Asia, the construction of sophisticated paddy rice response models and the substantial sample set can be quite difficult, because of the diverse phonologies of different paddy rice cultivation practices. Targeting this problem, this paper intended to explore the consistency in time series SAR features of different paddy rice regions. Two tropical paddy rice producing regions, Hainan Province of China and the Mekong River Delta of Vietnam, were selected as the study areas. Through the analysis of temporal responses using time series Sentinel-1 data, the behavior of paddy rice fields on time-series SAR were investigated and the comparisons of these two study areas were also conducted, which provided an evaluation for the portability of intelligent paddy rice mapping model in Southeast Asia.

The scattering features of synthetic aperture radar (SAR) object images are highly sensitive to different azimuth angles as well as attitudes, and the detection of azimuth-sensitive objects in SAR images becomes a challenging task in complex scenarios. In this paper, an improved YOLO v5-based azimuth-sensitive object detection method is proposed for such objects in SAR images that are azimuth-sensitive and of different scales. Firstly, the samples are grouped according to the scattering characteristics of objects in SAR images, and then the inverted residual (IR) structure incorporating the Squeeze-and-Excitation (SE) attention structure is introduced into the backbone network of YOLO v5 to improve the feature extraction capability for azimuth-sensitive objects. Taking aircraft in GF-3 1m SAR image as an example, the experiments show that the method has significantly improved the detection capability for azimuth-sensitive objects such as aircraft in SAR images, with a detection rate of 89.25% for the test imagery.

With the development of urbanisation in China, the urban areas are expanding rapidly, but there is a huge regional disparity between the east, central and western regions. The urban development in the western region lags far behind that in the eastern and central regions. In the western region of China, due to the large number of mountains and SAR backscatter mechanism, there are a lot of overlays in the image, resulting in high false alarms in built-up areas segmentation. In order to solve the problem, this paper proposed a new built-up area extraction model based on the Transformer. Different from the segmentation method based on convolutional neural network, the self-attention mechanism of the Transformer was introduced to effectively capture the image context information and reduce the impact of mountain overlays on the extraction of built-up areas. The multi-layer Transformer encoder and the multilayer perceptron (MLP) decoder were used to fuse feature maps of different scales for the sake of enhancing the ability to extract architectural features. With the purpose of improving the generalization ability, various data augmentation methods were used during training, such as random noise, random blur, and random distortion. In this paper, about 32000 samples, including some mountainous areas and around China, were used for training, which are from 27 scenes of GF-3 10m SAR images covering different areas in China. The method proposed in this paper reached a mIoU of 0.8130 and a Kappa coefficient of 0.9423, which significantly reduced false alarms in mountainous areas. Taking the study area of Lanzhou City, Gansu Province of China as an example, the result is basically consistent with the classification map of World Cover. It shows that the proposed method has a good ability to extract the distribution information of built-up areas.

As one of the most essential economic and industrial crops globally, corn holds a very important position in China's agricultural industry. Corn mapping is one of the most concerned fields in agricultural surveillance. However, compared with the utilization of backscattering coefficients, the polarimetric information was not fully discussed in previous corn mapping researches. In this paper, we use the coherency matrix of mid to late term multi-temporal fully polarimetric synthetic aperture radar (FP SAR) data to discriminate corn cultivation areas. The tensor representation is adopted for PolSAR analysis, with the help of multilinear principal component analysis (MPCA) to reduce feature dimensions. The importance of polarimetric information is discussed. This paper illustrates that good corn discrimination could be achieved with only mid to late term FP SAR data.

This paper displays a case study which accomplishes crop classification of simulated compact polarimetric (CP) SAR and fully polarimetric (FP) SAR images with Random Forest. Since the potential of CP SAR in classification has been illustrated by various researches, we intend to find out which of the polarimetric features are more superior in crop classification, through the importance rank of Random Forest classifier. Experiments are carried out based on an L-band AIRSAR FP SAR image and an ALOS-2/PALSAR-2 SM-2 FP image. Comparison analysis of feature importance between CP and FP demonstrates the intrinsic connotations of selected polarimetric characteristics, which provide guiding information for future investigation of CP SAR classification.

To explore the capability of quad polarimetric SAR images from a new pursuit monostatic mode of TerraSAR-X and TanDEM-X, a novel process chain for ship detection and velocity estimation is proposed in this paper. Compared with the classic processing chain used for single SAR image, more efficient techniques are integrated into this novel chain, which take advantage of the new image mode and its excellent imaging quality in term of the radiometric and geometric accuracies. Land masking based on coherence optimization and ship velocity estimation depended on the difference of positions from individual images are proposed as significant improvements in the new process chain. Their efficiency is validated with the experimental SAR images acquired over East China Sea. The experimental results show that the land regions and small islands are entirely removed and the estimation results of ship velocity is close to the ground truth from automatic identification system (AIS) data.