The detection and recognition of ships hold significant practical implications for both military and civilian departments. Recent advancements in deep learning technology have led to notable progress in this field. However, the precise detection and recognition of ships remains a challenge, especially in multi-source remote sensing images, due to their different feature expression and resolution. Moreover, the effectiveness of the existing models in complex environments still needs to be further validated. Therefore, in this letter, we construct a new ship dataset which contains five distinct ship categories under a number of complex environments. Different from existing datasets that are based on unimodal data, the new dataset uses multi-modal remote sensing images with different resolutions. On this basis, this study introduces an adaptable and robust ship detection and recognition model, namely SIDE-YOLO. The model incorporates an super-resolution convolutional neural network (SRCNN) and side window (SRSW) based contour feature enhancement module, a SimAM feature attention Resblock (ResBlockSA), and a DConv-based cross-scale feature enhancement block (DCFB), to strengthen the ship edge features, adaptively improve the problem of limited sample in SAR images, and amalgamate multi-scale ship context information, respectively. Validation results demonstrate that the proposed model achieves an average precision (AP) rate [mean AP (mAP)] of 84.31%, surpassing state-of-the-art ship recognition models. Notably, for the civilian ship category, the model exhibits 2.17% and 2.63% higher recall and AP in complex scenes, respectively.

Multi-modal remote sensing images (MRSI) often suffer from severe nonlinear radiation distortions (NRD) and significant geometric distortions, making precise matching challenging. We developed a feature-based matching algorithm to address this issue using edge consistency scale-space and significant relative response (ECSS). By designing an edge consistency filtering (ECF), we construct a scale space that preserves the structural information of MRSI at various scales, enhancing scale invariance. ECSS computes feature descriptors using multi-orientation filtering techniques to construct significant relative responses. This approach not only resists NRD but also utilizes information from all directional filters to build descriptors with higher discriminative power compared to direct filter responses or the maximum index map (MIM). To ensure rotational invariance, ECSS employs a robust technique for estimating the primary orientation. To further optimize matching accuracy and increase the number of effective matching points, ECSS uses a coarse-to-fine matching strategy. This involves using preliminary matching results to estimate the affine transformation between images, which then guides a more refined secondary matching. We evaluated the performance of ECSS on five different MRSI datasets and compared the results with nine state-of-the-art matching methods: ReDFeat, MINIMA-LG, RIFT, MS-HLMO, SRIF, WSSF, POS-GIFT, OFM, and GLS-MIFT. The experimental results demonstrate that ECSS excels in all performance metrics, particularly in terms of stability and matching accuracy when handling MRSI data with high NRD and complex geometric transformations.

Digital Surface Models (DSMs) extracted from multi-view satellite images have extensive applications in the filed of photogrammetry. Although Neural Radiance Fields (NeRF) has shown significant potential in 3D reconstruction, most existing NeRF methods for satellite scenes adopt an end-to-end single-branch network structure, making it difficult to achieve fine-grained modeling of multiple complex terrain simultaneously, and performs poorly in weak-texture regions. Meanwhile, deep MLP structures are prone to information degradation during feature transmission, further affecting the completeness and accuracy of DSMs. To address these challenges, we propose RDS-NeRF, a novel NeRF framework integrating residual feature enhancement and depth supervision. The method introduces a residual feature enhancement structure to alleviate the problem of information degradation during feature transmission in the network and improve the model’s ability to model local details and low-texture regions. Additionally, estimated depth maps are incorporated as global geometric priors to guide the network in constructing more accurate and complete 3D structures. Experiments on the WorldView-3 satellite imagery datasets across multiple typical land cover types (building, road, water body, and vegetation) and complex scenes integrating multiple land features demonstrate that RDS-NeRF outperforms mainstream methods in terms of DSM accuracy, completeness, and novel view synthesis quality. Ablation experiments further validate the complementarity and effectiveness of the residual enhancement and depth supervision mechanisms across different scene types. In conclusion, RDS-NeRF provides a new and effective solution for generating high-quality DSMs from satellite imagery with adaptability to multiple scenes. Code will be available at https://github.com/dfsvdgf/RDS-NeRF.

Conventional pixel-level seamline detection algorithms exhibit exponential time complexity on large, batch-mode remote-sensing mosaics, making it difficult to achieve an optimal trade-off between accuracy and efficiency. This paper introduces a globally optimal and highly efficient seamline detection framework. First, a preliminary seamline network is generated by iteratively clipping valid orthoimage regions with a Voronoi diagram, and image blocks are extracted only within overlap areas to markedly reduce data volume. Second, a cost graph constructed on down-sampled blocks is traversed in a reverse-diagonal Z-pattern; a “local entropy–gradient” composite cost function is applied, and a linear-time dynamic-programming (DP) scheme rapidly produces coarse seamlines that bypass texture-rich regions and confine the search space to a narrow band. Third, a buffer centered on the coarse seamline is created, within which an enhanced Dijkstra algorithm performs pixel-level refinement to accurately avoid complex obstacles. Experiments on the GF-7 data set demonstrate that, compared with five representative methods—SMP-DP, A*, Dijkstra, graph-cut, and OrthoVista—the proposed approach improves geometric accuracy by 14.46%, 58.69%, 50.20%, 17.79%, and 69.30%, respectively; processing efficiency is increased by 12.74%, 19.19%, 49.89%, >500%, and 83.72%, respectively. The algorithm has successfully mosaicked 627 GF-7 scenes covering the entire Henan Province, and has yielded similarly favorable results on ZY-3, GF-1 and GF-3 imagery, underscoring its high applicability and robustness for multi-source, large-format remote-sensing production.

Deep learning-based feature matching methods have been actively explored for satellite image bundle block adjustment, leveraging their inherent robustness to large geometric distortions and radiometric differences—key challenges in multi-source optical image processing. However, their practical utility remains limited by two critical bottlenecks: poor adaptability to extreme scale variations and prohibitive computational costs for large-format satellite data. To address these limitations, we propose a learning-based feature representation method enhanced by nonlinear diffusion filtering, with two targeted innovations: (1) Nonlinear diffusion filtering with terrain-adaptive parameters is integrated into a novel image tiling strategy, which preserves local feature integrity while enabling consistent correspondence across heterogeneous satellite data sources; (2) A top-down pyramid construction mechanism that incorporates local continuity constraints and an adaptive matching strategy selection protocol optimizes scale space exploration efficiency while safeguarding matching quality. Experiments on Earth observation and Martian image datasets validate the method’s superiority: it achieves the highest matching success rate (97.25%) and the highest BBA accuracy(1.65 pixel) among competing approaches, alongside competitive efficiency. This performance advantage is particularly pronounced under extreme scale differences and challenging imaging conditions, confirming its suitability for high-precision remote sensing applications.

GNSS-R technology for the retrieval of sea surface wind speed (SW) has gradually matured, and many research results in terms of methodology and accuracy have been obtained. Multisource data fusion has been a major trend in remote sensing research in recent years. However, there are few fusion algorithms in SW retrieval, and most of them retrieve the SW of a single data source. Based on the principle of CYGNSS forward scattering and HY-2B microwave scatterometer (HSCAT-B) backscattering, this paper proposes a Fusion Model (FM) of SW based on CYGNSS and HSCAT-B. For CYGNSS SW inversion using the FM, there is no need to input HSCAT-B data, and the accuracy of CYGNSS SW inversion above 10 m/s is improved. Based on the true SW data of the European Center for Medium-Range Weather Forecasts (ECMWF), the root mean square error (RMSE) of SW inversion is improved from 2.517 m/s and 1.645 m/s with a single data source to 1.527 m/s with the FM. To further correct the outliers of the FM, the result fitting model is added after the FM. The experimental results show that the RMSE of the result fitting model is improved from 1.527 m/s for the FM to 1.489 m/s. Finally, CYGNSS L2 SW and the National Data Buoy Center (NDBC) data is used to verify the inversion results, the RMSE of the result fitting model is 1.688 m/s and 1.60 m/s, respectively. The results prove the feasibility of a fusion algorithm for SW using multisource data.

Traditional deep learning-based object detection networks often resize images during the data preprocessing stage to achieve a uniform size and scale in the feature map. Resizing is done to facilitate model propagation and fully connected classification. However, resizing inevitably leads to object deformation and loss of valuable information in the images. This drawback becomes particularly pronounced for tiny objects like distribution towers with linear shapes and few pixels. To address this issue, we propose abandoning the resizing operation. Instead, we introduce Positional-Encoding Multi-head Criss-Cross Attention. This allows the model to capture contextual information and learn from multiple representation subspaces, effectively enriching the semantics of distribution towers. Additionally, we enhance Spatial Pyramid Pooling by reshaping three pooled feature maps into a new unified one while also reducing the computational burden. This approach allows images of different sizes and scales to generate feature maps with uniform dimensions and can be employed in feature map propagation. Our SCAResNet incorporates these aforementioned improvements into the backbone network ResNet. We evaluated our SCAResNet using the Electric Transmission and Distribution Infrastructure Imagery dataset from Duke University. Without any additional tricks, we employed various object detection models with Gaussian Receptive Field based Label Assignment as the baseline. When incorporating the SCAResNet into the baseline model, we achieved a 2.1% improvement in mAPs. This demonstrates the advantages of our SCAResNet in detecting transmission and distribution towers and its value in tiny object detection. The source code is available at https://github.com/LisavilaLee/SCAResNet_mmdet.

Highway crack segmentation is a critical task for highway infrastructure monitoring and maintenance. While imagery from unmanned aerial vehicles (UAVs) is applied to the task of highway crack segmentation, it has great prospects in terms of speed and range. However, it is difficult to accurately identify road cracks from UAV remote sensing images, because the cracks are very narrow and small, often containing only a few pixels. To improve the segmentation of road cracks in UAV images, this study proposed an improved identification technique based on the U-Net architecture enhanced with a convolutional block attention module, an improved encoder, and the strategy of fusing long and short skip connections. A public road crack dataset was relabelled for network training and a UAV remote sensing road crack dataset containing 1157 images was used to verify the generalization ability of the enhanced network model. Results showed that the proposed method could effectively predict highway cracks in UAV images, with mean intersection over union (mIoU) of 77.47% and crack accuracy of 68.38%, which was better than the traditional U-Net model and some traditional semantic segmentation models. The proposed network is trained quickly by public dataset and can predict the road cracks on the new UAV images with high crack accuracy. This study provides an effective solution for the need to quickly grasp the damage status of roads over a wide area in the case of earthquake and other natural disasters. The highway crack segmentation benchmark dataset has been open sourced at:https://github.com/zhhongsh/UAV-Benchmark-Dataset-for-Highway-Crack-Segmentation.

Using Cyclone Global Navigation Satellite System (CYGNSS) L1 data with large amount and wide coverage, this paper establishes a sea surface wind direction retrieval model based on three machine learning algorithms. Wind direction will cause the asymmetry of Delay Doppler Map (DDM). Based on this, this paper extracts two angle characteristic parameters from DDM. Compared with CYGNSS full DDM, L1 compact DDM has a reduced dimension. Therefore, this paper expands more characteristic parameters, including L1 parameters and geophysical parameters such as wind speed, mean sea surface pressure (MSL), sea surface temperature (SST). Wind speed, direction, MSL and SST are from the European Centre for Medium-Range Weather Forecasts (ECMWF). After data preprocessing, the experimental data set is generated. Based on this data set, this paper establishes SVM, BP and CNN wind direction retrieval models and verifies their model performance and generalization performance. In addition, a filter that can optimize the accuracy of CNN model is constructed, and the retrieval effect under different wind direction intervals is further studied. The results show that the accuracy of CNN model for L1 data is higher than that of SVM and BP, and the retrieval error of global sea surface wind direction after filtering is less than 20°. The accuracy difference of different wind direction intervals also has a significant impact on the wind direction retrieval results.

This letter is dedicated to evaluating the potential use of reflected signals from the BeiDou Navigation Satellite System for retrieving the thickness of sea ice. Accurate phase altimetry results over sea ice would be helpful for detecting the sea ice thickness. Here, a new phase processing approach is proposed to estimate the adaptable phase altimetry result. During the data processing, fake wrap points in the residual interferometric phase are replaced by fit values; moreover, the phase altimetry result is shown to reach centimeter accuracy. A coastal experiment was performed from January 22 to February 19, 2016, in Dashentang, Tianjin, China. The sea ice thicknesses calculated from the phase altimetry results and Archimedes' principle range from 5 to 25 cm, which are consistent with the thicknesses of samples collected during the experimental period in Bohai Bay near Dashentang.

With the development of the Chinese BeiDou satellite navigation system, the applications of BeiDou reflected (BeiDou-R) signals would play a key role in Global Navigation Satellite System reflected signals. Different from other navigation systems, the BeiDou satellite navigation system has certain unique characteristics, and the Geostationary Earth Orbit (GEO) satellite is one of them. The aim of this letter is to prove the feasibility of coastal ocean phase altimetry using BeiDou GEO reflected signals. The coastal experiment was performed from October 18, 2014, to October 19, 2014, in Dayang Shan, Zhejiang, China. This is the first coastal ocean phase altimetry experiment using BeiDou GEO reflected signals. The phase altimetry results can invert the slope of ocean surface height variation (in time), and its highest accuracy can reach centimeter level.

The paper explored a method to obtain accurate lake surface heights using Global Navigation Satellite System (GNSS) Coarse/Acquisition Code (C/A)reflected from the ocean surface. The method is referred to as Global Navigation Satellite System-Reflection (GNSS-R) (C/A) code altimetry. It focuses on the extraction of the delay between the direct and reflected signal and inverting ocean altimetry with the delay and the geometric relationship. The ocean altimetry results are consistent with the height results of the differential positioning analysis of the data collected by a handheld GIS. The results show that we can achieve meter level height in seven minutes average.