Multisource remote sensing images (RSIs) can capture the complementary information of ground objects for use in semantic segmentation. However, there can be inconsistency and interference noise among the multimodal data from different sensors. Therefore, it is a challenge to effectively reduce the differences and noise between the different modalities and fully utilize their complementary features. In this article, we propose a universal cross-fusion transformer framework (CFFormer) for the semantic segmentation of multisource RSIs, adopting a parallel dual-stream structure to extract features separately from the different modalities. We introduce a feature correction module (FCM) that corrects the features of the current modality by combining features from the other modalities in both the spatial and channel dimensions. In the feature fusion module (FFM), we employ a multihead cross-attention mechanism to interact globally and fuse features from the different modalities, enabling the comprehensive utilization of the complementary information in multisource RSIs. Finally, comparative experiments demonstrate that the proposed CFFormer framework not only achieves state-of-the-art (SOTA) accuracy but also exhibits outstanding robustness when compared to the current advanced networks for semantic segmentation of multisource RSIs. Specifically, CFFormer achieves a mean intersection over union (mIoU) of 58% and an overall accuracy (OA) of 85.35% on the WHU-OPT-SAR dataset, outperforming the second-ranked network by 4.71% and 1.74%, respectively. On the Vaihingen and Potsdam datasets, CFFormer also achieves the best results, with mIoU and OA values of 84.31%/91.88% and 88.62%/92.64%, respectively. The source code is available athttps://github.com/masurq/CFFormer.

Efficient and accurate wetland monitoring is of great significance for controlling climate, preventing floods, and maintaining ecological balance. Due to the characteristics of complex wetland features, there are still problems in feature extraction and classifier selection when dealing with wetland mapping using remote sensing data. In this paper, a novel wetland classification method based on Convolutional Neural Network (CNN) and Support Vector Machine (SVM) is proposed. Firstly, multi-source images are constructed by sentinel-1 and 2. Furthermore, deep features are extracted from multi-source images based on pre-trained CNN. Finally, considering the advantages of SVM in remote sensing classification, the softmax is replaced by L2-SVM. To verify the effectiveness of the proposed method, Qilihai Wetland is used in our experiment. Experimental results show that combining multi-source remote sensing images significantly improves wetland classification accuracy. Moreover, the proposed method has superior performance with OA and Kappa of up to 90.3% and 0.870, especially in small sample categories and complex land-cover types.

As one of the most important means for earth observation, synthetic aperture radar (SAR) ship detection has been playing an increasingly important role recently. However, the accuracy of one-stage detectors is relatively low when ships are densely arranged in SAR images. It means that current detectors are insufficient to meet the application requirements. In this paper, an improved YOLOv5 detector, which is based on the attention mechanism and upsampling operator, is proposed. We adopt the ECA attention mechanism to solve the issue of insufficient feature extraction capability for ship targets. Besides, we propose the CARAFE upsampling operator to enhance the proportion of ship detail information in reconstructed feature maps. Experiments on the SAR ship detection dataset (SSDD) demonstrate that the detection P, R, and mAP metrics have improved by 4.1%, 4.8%, and 2% compared with the original YOLOv5 algorithm.

Synthetic Aperture Radar (SAR) has the ability of all-weather and all-day observation, which is suitable for flood monitoring. However, there still has some challenges in flood monitoring using SAR images, such as high-quality prior knowledge, accumulated errors in time series analysis, and the impact of other land cover changes. To solve these problems, in this paper, a novel unsupervised flood monitoring method using temporal information and statistical characteristics of SAR images is proposed. Firstly, the temporal-spatial-polarization (TSP) dataset is constructed from different SAR data. Furthermore, improved K-means clustering is proposed to fit these constructed datasets to mitigate the error accumulation. Finally, considering the statistical characteristics of SAR data, the Bray-Curtis distance is applied to optimize improved K-means. To verify the effectiveness of the proposed method, the latest flood event in Jingpo Lake in China with temporal Sentinel-1 data is used. The experimental results demonstrate that our method has superior performance in detecting flood regions, with OA and Kappa of up to 97.64% and 0.86.

Synthetic Aperture Radar (SAR) is vital for coastal wetland mapping due to its cloud and vegetation penetration capabilities. However, using SAR imagery in wetland mapping is challenged by similar backscatter coefficients and speckle noise. To address these issues, a patch-based cascade forest (PBC) method is proposed in this paper, which combines multi-temporal, full-polarization GF-3 SAR data to map typical ground objects in the Yellow River Delta wetlands. Comparative experiments are conducted using different temporal datasets, and the proposed method is compared with Random Forest (RF), Support Vector Machines (SVM), eXtreme Gradient Boosting (XGB), and multi-Grained Cascade Forest (gc-Forest). The results show that the multi-temporal patch-based cascade forest method, achieving an Overall Accuracy (OA) of 89.46% and a Kappa coefficient of 0.872, significantly outperforms single-temporal methods and other machine learning algorithms, which proved that our method fit for long-term, high-quality wetland classification.

Efficient classification is crucial to mastering wetland land cover types and facilitating conservation. Due to the advantages of Synthetic Aperture Radar (SAR) imaging, it enables continuous monitoring and penetration through vegetation canopies. In general, wetland types in SAR imagery mainly rely on backscattering, which hard to distinguish and differentiate land cover using single features. Effectively leveraging SAR features can improve classification accuracy in wetlands. However, there is rarely research discussing feature effects in classification. In this paper, physical and data-driven-based feature extraction methods are analyzed in wetland classification. Firstly, physical and data-driven models are combined for feature extraction. Furthermore, the performance of common classifiers is compared to evaluate the effectiveness of different feature types, such as Support Vector Machine (SVM), Random Forest(RF), and Extreme Gradient Boosting(XGB). The experimental results indicate that the combined feature extraction method of the physical and data-driven model performs the best in classifying the Yellow River Delta. It also improves the accuracy of all the classifiers in the comparative experiment. Notably, the RF classifier achieves the highest classification accuracy, with an overall accuracy(OA) of 94% and a Kappa coefficient of 0.93.

A new flood extraction method which combines semi-automatic thresholding and change detection is proposed. First, a line across land and water is drawn manually. The locally optimal threshold is calculated automatically along the line from two endpoints to middle. Using this threshold, the low backscattering regions are extracted to generate a preliminary flood map. Then, the neighborhood-based change detection method combined with the entropy thresholding is adopted to detect the changed area. Finally, pixels in both the low backscattering regions and the changed regions are marked as “flood”. The effectiveness and practicality of the flood extraction method was demonstrated by a set of Sentinel-1A data and ground truth data provided by the Copernicus Emergency Management Service (EMS).

Long-term continuous monitoring of coal mining activities is conducive to master the development and potential risk factors of the mining area. Radar reflected echo signal contains the intensity and phase information, which is suitable for detecting different types of mining areas. However, relevant coal mining detection research mainly focuses on phase information, which results in the application based on underground coal mines more than open-pit coal mines. In order to take full advantage of Synthetic Aperture Radar (SAR) data, in this paper, a novel mine detection method based on intensity and phase information is introduced to detect the underground and open pit coal mine using the multi-temporal ALOS data in Shenmu County, China. Firstly, Interferometric SAR (InSAR) technology is used to detect the deformation information caused by underground coal mine. Secondly, the coherence information is used to detect the deformation from both underground and open-pit coal mine. Moreover, change detection method based on intensity information is used to detect the open-pit coal mining. Finally, the results of phase, coherence and intensity are combined to detect the underground and open-pit coal mine. The experimental results of Shengdong mine show the effectiveness of the proposed method.

Hyperspectral/multispectral and panchromatic of optical remote sensing images are commonly used for multisensor image fusion, which has been applied in various applications of Earth observation. However, the utilization of optical remote sensing data suffers from the limitation of bad weather and cloud contamination. To address aforementioned issue and enhance spatial details of polarimetric synthetic aperture radar (PolSAR) image, a novel dual-domain super-resolution image fusion method is proposed by combining improved spherically invariant random vector (ISIRV) model with generalized adaptive linear combination approximation (GALCA) technology in this study. The proposed method decomposes the task of image fusion into polarimetric and texture domain fusion by integrating polarimetric components of PolSAR image and texture detail component of panchromatic image, which can significantly improve spatial resolutions of the PolSAR image while preserving polarimetric information. The data fusion experiment is implemented with three data sets including panchromatic images of Gaofen-1 (GF-1) and Gaofen-2 (GF-2) and the quad-pol SAR data of Gaofen-3 (GF-3) and Radarsat-2. Results show that the proposed dual-domain image fusion method provides a better performance compared with state-of-the-art multisensor fusion methods (BT, PCA, GS, indusion, and PRACS) regarding qualitative and quantitative evaluations. In addition, results of image fusion are applied to image classification over agricultural and urban areas of China, which shows that classification accuracy is significantly improved when compared with the result using only the original image.

A modified model-based polarimetric decomposition method, considering both the surface and dihedral scattering depolarization, is proposed for soil moisture retrieval. In the parameter solution, it works combining at least two polarimetric SAR images acquired simultaneously in different incidence angles. Moreover, it needs not to decide whether dihedral or surface scatter is the dominant contribution. The experiments to demonstrate the potential of the proposed approach is carried out using L-band polarimetric UAVSAR multi-incidence data in Winnipeg, Canada. The scattering mechanism of forest, grass land, urban, and barren area are analyzed compared with Yamaguchi three-component decomposition results. The performance of the soil moisture estimation algorithm is also assessed by comparing the retrieval results with in situ measurements.

Accurate and timely change detection of Earth's surface features is extremely important for understanding relationships and interactions between people and natural phenomena. Post-Classification Comparison (PCC) methods based on supervised change detection are widely used in change detection for remote sensing images, but are easily affected by a significant cumulative error of single remote sensing image classification. Unsupervised change detection methods are affected by the speckle noise and cannot explicitly identify the types of land cover or land use transitions. To solve those problems, this paper proposes a change detection method based on similarity measure and joint classification. The similarity measure is obtained by test statistic and Kittler and Illingworth minimum-error thresholding algorithm (TSKI), which is used to automatically control the joint-classification classifier. The efficiency of the proposed method is demonstrated by the polarimetric synthetic aperture radar (PolSAR) images acquired by Radarsat-2 over Wuhan of China. The experimental results show that the method can identify different types of land cover changes and reduce the false alarms in the change detection.

A lodged wheat detection algorithm by applying a false alarm rate to the circular-pol correlation coefficient (CCC) and the total scattered power (Span) is proposed. The CCC is first used to identify non-lodged wheat, which is reflection symmetry. The Span feature is introduced to distinguish lodged wheat from canola in the study area, according to their large difference in scattering intensity. The Polarimertric synthetic aperture radar (PolSAR) image acquired by the Radarsat-2 satellite over the Yigen farmland of China, was used to validate the effectiveness of the proposed approach. The results indicate the potential of using only post-event PolSAR image to detect the lodged wheat.

This study investigates the characteristics of periodic surface in agriculture by the use of C-band polarimetric synthetic aperture radar (PolSAR) imagery. The scattering characteristics of the periodic potato fields in different directions are highlighted using a set of polarimetric parameters. Enhanced coherent scattering is observed when the alignment direction of ridging patterns is perpendicular to radar's line of sight (LOS). There are higher copolarized backscattering coefficients and unaffected cross-polarized backscattering coefficient for the coherent scattering. The increased copolarized correlation coefficient, reduced entropy and polarimetric alpha angle indicate that the induced coherent scattering has small scattering randomness and the odd scattering is its dominant scattering mechanism.