Marine oil spills pose significant threats to ecosystems and human health, emphasizing the importance of synthetic aperture radar (SAR) images for reliable and all-weather monitoring. However, current methods face two major challenges. The first is data limitations, including insufficient data quantity and noise, such as speckle noise and distortions introduced during preprocessing. The second is look-alike confusions, which pose challenges in distinguishing oil spills from visually similar phenomena. This article introduces TransOilSeg, a novel method designed to address these challenges and enhance oil spill detection performance. TransOilSeg employs a transfer learning component (TLC) to integrate data from diverse geographical regions and varying quality, learning general features from multisource datasets. By leveraging a gradient aggregation algorithm, the model combines features from limited and noisy SAR oil spill (SOS) datasets, transferring data deficiencies. In addition, the adaptive attention hybrid encoder (AAHE) analyzes contextual features and adapts to varying datasets, enabling the model to effectively distinguish oil spills from look-alike phenomena. Comprehensive evaluations across multiple datasets demonstrate the robust generalization capability of TransOilSeg. On the M4D dataset, which includes 1002 training samples, the model achieved a mean intersection over union (mIoU) of 61.38% for oil spill detection and 62.41% for look-alike detection. Furthermore, TransOilSeg maintained strong performance when transferred between datasets with varying levels of noise and distortions, demonstrating its adaptability to challenging conditions. These results highlight its potential as a reliable tool for marine oil spill detection and monitoring.

Ship target search tasks aim to match specific ships across two or more satellite images. Like pedestrian and vehicle re-identification tasks in computer vision, accurate ship re-identification encounters challenges, including subtle differences between ships of the same type and substantial intra-instance variations due to satellite angle of view and spectral differences. To tackle these challenges, this paper introduces a deep learning-based two-branch framework for ship target search, integrating ship detection and re-identification tasks. One branch extracts the target ship features while the other captures the search region features. These features are then fused through a dedicated layer, and the final output is derived from the keypoint detection header. A new dataset was curated using Sentinel-2 and Gaofen-1 satellite data. Experimental results validate the robustness of the proposed method, achieving an accuracy of 94.37% on the new dataset. Our method’s scalability has been validated through experiments using CBERS-04 and Gaofen-6 satellite data.

Waterline extraction from synthetic aperture radar (SAR) images can be transformed into a sea-land segmentation task. However, two aspects of deep learning sea-land segmentation have been ignored in the literature: 1) deep learning models are commonly built using whole images rather than focusing on their sea-land transition parts and 2) a higher resolution input may not render a better segmentation result under the constraint of a fixed-size receptive field. Our investigation on the aspects indicates that focusing the modeling process on the sea-land transition parts can benefit the waterline extraction, and the highest resolution may not be the best choice for all pixels. We proposed masked soft intersection over union (MSIoU) loss and the difference-focusing fusion decision (DFFD) ensemble learning method. MSIoU loss incorporates the mask of the transition parts to focus the modeling process on the transition parts. The DFFD ensemble learning method imitates manual labeling and can avoid selecting the resolution of input images. The DFFD ensemble model’s member models segment an image’s sea and land areas at different resolutions. Then, its fusion model further recognizes the pixels where the member models inconsistently predict sea-land types. We applied the DFFD ensemble model to 10-m-resolution SAR images of the test set. Compared to the traditional single-resolution model, the DFFD ensemble model with MSIoU loss achieved a 10.32–12.58-m accuracy in waterline extraction with a 2.08–2.78-m error reduction in the study area. Moreover, the DFFD framework is independent of data and model choice and can be readily modified for other segmentation tasks.

Synthetic aperture radar (SAR) can monitor the sea surface imprints of tropical cyclones (TCs) with high spatial resolution, day and night. Automatically locating TC center positions in SAR images is a challenging task. This article developed a two-stage, fully automatic TC-center estimation algorithm. First, the sea surface wind directions (SSWDs) at SSWD points are retrieved by the improved local gradient (ILG) method. We incrementally deflected the SSWD outward at a 0.5° angle from −50° to 10° (the negative angles represent clockwise deflection). The heat maps are generated for each of the 121 angles, and the values at each heat map are the cumulative numbers of the lines perpendicular to the compensated SSWDs. The site corresponding to the maximum cumulative number in all 121 heat maps is the coarsely estimated center position. This center search is the culmination if it falls outside the SAR image. Otherwise, the second stage is triggered, and the sub-SAR image (150 km$\times150$km) centered at the coarsely estimated center position is extracted. Then, the first-stage procedure is repeated with the sub-SAR image to precisely estimate the center position. Optionally, the precisely estimated center position can be further adjusted by considering that normalized radar cross section (NRCS) is normally minimal at the TC center. We applied the algorithm to 87 SAR images. Five of these images do not contain TC centers. The results are in good agreement with the visually located TC center positions and those in the best track (BT) datasets.

Wind-induced oriented textures (WIOTs) are commonly used to retrieve sea surface wind directions from synthetic aperture radar (SAR) images. In this study, we found that WIOTs are also related to sea surface wind speeds (SSWSs). The entropy values in the gray-level cooccurrence matrices (GLCMs) for SAR images containing WIOTs will become steady with increasing distance between pairs of pixels. Furthermore, these steady values of entropy (SVEs) show a clear linear relationship with SSWSs. As a result, an SSWS retrieval model was developed based on this relationship. We used 2222/2223 Sentinel-1 SAR images (wind speed ranges from 5 to 20 m/s) to fit/validate the algorithm. The retrieved SSWSs were compared with the European Centre for Medium-Range Weather Forecast (ECMWF) SSWSs, Cross-Calibrated Multi-Platform (CCMP) SSWSs, and Tropical Atmosphere/Ocean (TAO) buoy measurements, and the root-mean-square differences (RMSDs) were 1.78, 1.70, and 1.78 m/s, respectively. The new model was also tested for SAR images acquired under hurricane conditions. The wind comparisons against stepped-frequency microwave radiometer (SFMR) measurements show an RMSD of 1.28 m/s. Our model’s performance was also tested with the images at different spatial scales in the validation data set. Since the model is based on inherent image patterns, it still works well for SAR images without precise calibration.

This paper assesses different retrieval schemes used for the Chinese Gaofen-3 Synthetic Aperture Radar (GF-3 SAR) co-polarized data. The data consist of 4186 GF-3 data points and collocated wind information from sources including the ASCAT scatterometer, HY2A-SCAT scatterometer, and National Data Buoy Center (NDBC) buoy wind data set. The VV-polarized geophysical model function (GMF) is a CMOD7 model while the HH-polarized GMF is a hybrid of the CMOD7 and PR model. Assessments involve comparisons between SAR-derived and collocated winds in terms of the root-mean-square difference (RMSD) and bias. First, a comparison between the two retrieval schemes for the VV-polarized data clearly shows that the optimal scheme performs better than the classical scheme for wind speed retrieval. Comparisons for HH-polarized data show similar results. These experiments indicate that the wind speed RMSDs for the GF-3 co-polarized data are within 2 m/s when using the optimal scheme. Moreover, the wind direction RMSDs from the two schemes have no significant difference, with values near 20°. Overall, these assessments indicate that the GF-3 co-polarized data are sufficient for operational wind speed retrieval using the optimal scheme. However, wind direction retrieval requires further improvement.

The brightness temperature data acquired during 2012-2015 from the scanning microwave radiometer (SMR), onboard the first Chinese ocean dynamic environment satellite- Haiyang-2A, were matched up with the WindSat Polarimetric Radiometer (WindSat) 0.25° × 0.25° gridded daily sea surface temperature (SST) data. Then, the artificial neural network (ANN) ensemble (ANNE) method implementing the Crogging technique was used to build the SMR SST retrieval algorithm. Different from a regular ANN, an ANNE combines the outputs of its ANN members to generate an algorithm. The developed ANNE algorithm for SMR SST was validated based on the SMR/WindSat data pairs that were not used in the tuning of the algorithm. The SST comparison shows the root mean square (rms) of 1.16 °C for the ANNE algorithm. We further validate the SMR SST products using the in situ measurements from the National Oceanic and Atmospheric Administration iQuam System. The rms of the ANNE algorithm in comparison with the global iQuam SSTs is 1.46 °C. All validations showed that ANNEs were more accurate than the other statistically based SST retrieval algorithms for SMR, and generally had much smaller uncertainties than regular ANNs.

A gray-level co-occurrence matrix (GLCM)-based method was developed for better texture orientation estimation in remote sensing imagery. A GLCM is essentially the joint probability distribution of gray levels at the position pairs satisfying a specific relative position within an image. We first found that when the relative position is aligned with texture orientation, larger elements of the corresponding GLCM are concentrated diagonally. Then, we developed a new texture orientation estimation method. The method uses the GLCMs of relative positions equally spaced in orientation and distance, and three schemes of these GLCMs are calculated. A GLCM-derived parameter is then defined to quantitatively measure the degree of diagonal concentration of the GLCM elements, and its integral over the variable of relative distance is selected as an indicator to find the dominant texture orientation(s). For testing, we applied the method to 44 selected images containing one or multiple aligned textures. The results show that the method is in good agreement with visual inspections from 45 randomly selected people, and is insensitive to large typical noises and illumination change. In addition, using (any) one GLCM calculation scheme over the others does not significantly affect the results. Finally, the method was applied to sea surface wind direction (SSWD) retrieval from 89 synthetic aperture radar images. In the application test, the developed method achieves better SSWD retrieval accuracy than do the commonly used Fourier transform- and gradient-based methods by 8.13° and 16.09° against the European Centre for Medium-Range Weather Forecast ERA-Interim reanalysis data and 10.21° and 17.31° against the cross-calibrated multiplatform data.

In this letter, we present a business process flow of marine target detection using GF-3 SAR data. It includes 12 kinds of imaging modes and 3 types of polarization. The detection method for the single polarization data is a Double-Parameter Constant False Alarm Ratio (DP-CFAR) algorithm. For double polarization data, a pixel level fusion strategy between two single polarization data is used. For the four polarization data, the method of target detection includes three steps: decomposition, false color composite, and supervised classification. All the data we used are test data of the GF-3 satellite. The results of target detection can be visually checked, and can also be validated by an automatic identification system (AIS), which will be done next month. This is submitted for a special session of “New Developments of Chinese Oceanographic and Meteorological Satellites”.

Underwater topography can be imaged using remote sensing methods. However, current-topography interaction constitutes the weakest link in the procedures of remote sensing imaging mechanism of underwater topography. Some researchers have studied the effect of simplified topography on a parallel or vertical tidal current field. In this research, we analyzed the modulation effect of skew topography (including sand ridge and channel) on unidirectional tidal current using a 3-dimensional (3-D) hydrodynamic model. Unidirectional current had different responses to different topography types. And the distribution of surface current velocity and velocity gradient showed opposite features over sand ridge and channel.

Remote sensing images often suffer low contrast. Although many contrast enhancement methods have been proposed in recent literature, the efficiency and robustness of remote sensing image contrast enhancement is still a challenge. In this letter, a novel self-adaptive histogram compacting transform-based contrast enhancement method for remote sensing images is presented to meet with the requirements of automation, robustness, and efficiency in applications. First, the histogram of an input image is optimized into compact and continuous status with the constraints of the merging cost, the moderate global brightness, and the entropy contribution of gray levels. Then, a local remapping algorithm is proposed to catch more details during the course of gray extending with the linear stretch. Finally, a dual-gamma transform is proposed to enhance the contrast in both bright and black areas. Experimental and comparison results demonstrate that the proposed method yields better results than the state-of-the-art methods and maintains robustness in different cases. It provides an effective approach for remote sensing image automatic contrast enhancement.

The algorithms discussed in this paper are three Constant False Alarm Rate (CFAR) models, which include the Probabilistic Neural Network(PNN) model, the K-Gamma model and the double parameters model. The SAR data utilized in the paper include ERS-2, ENVISAT and Radarsat SAR data. The data are applied in ship detection experiments and the results of ship detection of three models are compared. The results show that the PNN model's applicability is the best .The performance of PNN model in ERS and ENVISAT SAR data is better than the K-Gamma model. The K-Gamma model can only do well in Radarsat SAR data. The double parameters model can fit local distribution of SAR image in the sea.

This paper presents an improved constant false alarm rate (CFAR) model for ship detection in synthetic aperture radar (SAR) imagery. The model includes the probabilistic neural networks, CFAR technique, golden section method and area growth method. It is compared with other ship detection methods. The results show that the improved CFAR model performs well