Sea ice is a significant factor in influencing environmental change on Earth. Monitoring sea ice is of major importance, and one of the main objectives of this monitoring is sea ice classification. Currently, synthetic aperture radar (SAR) data are primarily used for sea ice classification, with a single polarization band or simple combinations of polarization bands being common choices. While much of the current research has focused on optimizing network structures to achieve high classification accuracy, which requires substantial training resources, we aim to extract more information from the SAR data for classification. Therefore we propose a multi-featured SAR sea ice classification method that combines polarization features calculated by polarization decomposition and spectrogram features calculated by joint time-frequency analysis (JTFA). We built a convolutional neural network (CNN) structure for learning the multi-features of sea ice, which combines spatial features and physical properties, including polarization and spectrogram features of sea ice. In this paper, we utilized ALOS PALSAR SLC data with HH, HV, VH, and VV, four types of polarization for the multi-featured sea ice classification method. We divided the sea ice into new ice (NI), first-year ice (FI), old ice (OI), deformed ice (DI), and open water (OW). Then, the accuracy calculation by confusion matrix and comparative analysis were carried out. Our experimental results demonstrate that the multi-feature method proposed in this paper can achieve high accuracy with a smaller data volume and computational effort. In the four scenes selected for validation, the overall accuracy could reach 95%, 91%, 96%, and 95%, respectively, which represents a significant improvement compared to the single-feature sea ice classification method.

Only approximately 20% of the global seafloor topography has been finely modeled. The rest either lacks data or its data are not accurate enough to meet practical requirements. On the one hand, the satellite altimeter has the advantages of large-scale and real-time observation. Therefore, it is widely used to measure bathymetry, the core of seafloor topography. However, there is often room to improve its precision. Multibeam sonar bathymetry is more precise but generally limited to a smaller coverage, so it is in a complementary relationship with the satellite-derived bathymetry. To combine the advantages of satellite altimetry-derived and multibeam sonar-derived bathymetry, we apply deep learning to perform multibeam sonar-based bathymetry correction for satellite altimetry bathymetry data. Specifically, we modify a pretrained VGGNet neural network model to train on three sets of bathymetry data from the West Pacific, Southern Ocean, and East Pacific. Experiments show that the correlation of bathymetry data before and after correction can reach a high level, with the performance of R² being as high as 0.81, and the normalized root-mean-square deviation (NRMSE) improved by over 19% compared with previous research. We then explore the relationship between R² and water depth and conclude that it varies at different depths. Thus, the terrain specificity is a factor that affects the precision of the correction. Finally, we apply the difference in water depth before and after the correction for evaluation and find that our method can improve by more than 17% compared with previous research. The results show that the VGGNet model can perform better correction to the bathymetry data. Hence, we provide a novel method for accurate modeling of the seafloor topography.

Side-scan sonar and multi-beam echo sounder (MBES) are the most widely used underwater surveying tools in marine mapping today. The MBES offers high accuracy in depth measurement but is limited by low imaging resolution due to beam density constraints. Conversely, side-scan sonar provides high-resolution backscatter intensity images but lacks precise positional information and often suffers from distortions. Thus, MBES and side-scan images complement each other in depth accuracy and imaging resolution. To obtain high-quality seafloor topography images in practice, matching between MBES and side-scan images is necessary. However, due to the significant differences in content and resolution between MBES depth images and side-scan backscatter images, they represent a typical example of heterogeneous images, making feature matching difficult with traditional image matching methods. To address this issue, this paper proposes a feature matching network based on the LoFTR algorithm, utilizing the intermediate layers of the ResNet-50 network to extract shared features between the two types of images. By leveraging self-attention and cross-attention mechanisms, the features of the MBES and side-scan images are combined, and a similarity matrix of the two modalities is calculated to achieve mutual matching. Experimental results show that, compared to traditional methods, the proposed model exhibits greater robustness to noise interference and effectively reduces noise. It also overcomes challenges, such as large nonlinear differences, significant geometric distortions, and high matching difficulty between the MBES and side-scan images, significantly improving the optimized image matching results. The matching error RMSE has been reduced to within six pixels, enabling the accurate matching of multi-beam and side-scan images.

Precision bottom tracking is a core step in the data processing of side scan sonar (SSS), which is of critical value for the quality of the final SSS production. Currently, the automatic precision SSS bottom tracking remains challenging due to the complex noise caused by the measuring environment, especially the existing methods did not fully exploit the temporal correlations or depended on the hand-crafted setting. Therefore, we proposed a novel temporal fusion based one-dimensional (1-D) sequence semantic segmentation model, TFSSM-1D, to fuse the temporal correlation features and perform automatic precision SSS bottom tracking. The TFSSM-1D uses the deep learning encoder-decoder model for mapping inputs to 1-D semantic label outputs, with the aid of preprocess and temporal fusion modules to improve the accuracy and robustness. Among them, preprocess module is used to introduce the prior knowledge of bilateral symmetry, which alleviates the defect of long-distance features correlation caused by the inductive bias of convolution neural network. The temporal fusion consists of the point-wise temporal fusion module (PTFM) and the bi-directional attention propagation module (BAPM) guides the model to explicitly fuse the temporal variation features on different scales. The experimental results demonstrate the effectiveness of TFSSM-1D, and its mean port offset error reaches 2.7058 on the testing set without down-sampling, which is 40% lower than the previous deep learning based model with single ping input, and other evaluation metrics have also significantly improved. The inference on unseen data shows that TFSSM-1D can achieve precision bottom tracking with good noise immunity.

Shallow bathymetry is important for ocean exploration, and the development of high-precision bathymetry inversion methods, especially for shallow waters with poor quality, is a major research aim. Synthetic aperture radar (SAR) image data benefit from a wide coverage, high measurement density, rapidity, and low consumption but are limited by low accuracy. Alternatively, multibeam data have low coverage and are difficult to obtain but have a high measurement accuracy. In this paper, taking advantage of the complementary properties, we use SAR image data as the content map and multibeam images as the migrated style map, applying the VGG19 neural network (optimizing the loss function formula) for bathymetric inversion. The model was universal and highly accurate for bathymetric inversion of shallow marine areas, such as turbid water in Taiwan. There was a strong correlation between bathymetric inversion data and measured data (R² = 0.8822; RMSE = 1.86 m). The relative error was refined by 9.22% over those of previous studies. Values for different bathymetric regions were extremely correlated in the region of 20–40 m. The newly developed approach is highly accurate over 20 m in the open ocean, providing an efficient, precise shallow bathymetry inversion method for complex hydrographic conditions.

Near-shore shallow-water bathymetry is critical for coastal management, with important scientific and engineering value. Compared to traditional bathymetric approaches, satellite-derived bathymetry (SDB) is more efficient and cost-effective. Especially, the Ice, Cloud and land Elevation Satellite 2 (ICESat-2) carries a photon-counting 532-nm laser capable of penetrating optically shallow waters for bathymetry without requiring in situ measurements. In practical application, heterogeneous benthic substrates and variable water optical properties pose significant out-of-distribution (OOD) challenges to SDB. However, current filtering or clustering models for ICESat-2 SDB, relying on constant configuration and narrow applicability ranges, are unadapted to complex geomorphological conditions. Here, we propose the photon density profile guided cross-modality fusion point cloud segmentation (DECOMPOSE) model for the ICESat-2 SDB mission. Unlike spatially limited conventional approaches, DECOMPOSE achieves comprehensive seafloor characterization by fusing global density profile distributions with microstructural point cloud details. It is resilient against OOD challenges and generalizable to a variety of environments. On a geographically isolated testing dataset, DECOMPOSE outperformed baseline models, with a mean intersection-over-union of 0.9559%, 15.2% higher than the suboptimal comparative model. On a real-world OOD dataset, the DECOMPOSE bathymetric accuracy was excellent, with a root mean square error (RMSE) of 0.4938 m, a mean absolute error (MAE) of 0.2760 m, and a goodness-of-fit R2 exceeding 0.98. To further improve the usefulness of the proposed paradigm, we integrated DECOMPOSE with a multispectral imagery-based SDB model to establish the data-driven active–passive fusion SDB (DAPF-SDB) model, designed to provide full-coverage underwater topography mapping without in situ data. Quantitative and qualitative analyses demonstrated that, relative to the baseline models, DAPF-SDB achieves accurate and robust seafloor topography mapping under OOD real-world conditions, reducing RMSE and MAE by more than 16% and 14%, respectively. Code for DECOMPOSE and DAPF-SDB is available at https://github.com/qxm1995716/decompose