Optical and synthetic aperture radar (SAR) image matching presents a formidable challenge due to their pronounced geometric and radiometric distinctions arising from multi-modality. The distinct imaging mechanisms of optical and SAR sensors make it challenging to identify essentially homologous points in the physical sense, raising concerns about the accuracy and repeatability of correspondences in current feature matching methods. In this study, we introduce a detector-free feature matching algorithm specifically designed to match optical and SAR images through a two-step strategy. In the initial phase, our proposed method conducts pixelwise matching using downsampled feature descriptors, eliminating the necessity to identify repeatable keypoints. To mitigate complexity, we enforce a pseudo-epipolar constraint to reduce computational costs by constraining the search range. Subsequently, refined matching is performed on the initial correspondences to rectify inaccuracies in the pixelwise matching localization of the first step. Both matching steps are implemented on a graphics processing unit (GPU) to ensure high efficiency. The proposed algorithm attains an average matching accuracy of 2.39 pixels and operates with an efficiency of 1.09 seconds for 1108 image pairs, underscoring its superior comprehensive performance compared to various state-of-the-art algorithms, including hand-crafted methods and deep-learning networks.

Stereo matching, a critical step of binocular 3D reconstruction, has fully shifted to deep learning due to its strong feature representation of remote sensing images. However, the ground truth for the stereo matching relies on expensive airborne LiDAR data, thus making it difficult to obtain enough samples for supervised learning. To improve the generalization ability of stereo matching networks on cross-domain data from different optical sensors and scenarios, in this paper, we are dedicated to studying the key training factors from three perspectives. (1) When selecting a training dataset, prioritize data with similar regional target distribution as the test set, rather than relying on data from the same sensor. (2) Regarding the training modes, unsupervised methods generalize better than supervised methods. (3) We devised an unsupervised early-stop strategy to help preserve the best model based on the pretrained weights. Extensive experiments are conducted to support the previous findings, on the basis of which we present an unsupervised stereo matching network with good generalization performance. We release the source code and the datasets at https://github.com/Elenairene/RKF RSSM to reproduce the results and encourage future work.

With the emergence of the Smart City concept, the rapid advancement of urban three-dimensional (3D) reconstruction becomes imperative. While current developments in the field of 3D reconstruction have enabled the generation of 3D products such as Digital Surface Models (DSM), challenges persist in accurately reconstructing shadows, handling occlusions, and addressing low-texture areas in very-high-resolution remote sensing images. These challenges often lead to difficulties in calculating satisfactory disparity maps using existing stereo matching methods, thereby reducing the accuracy of 3D reconstruction. This issue is particularly pronounced in urban scenes, which contain numerous super high-rise and densely distributed buildings, resulting in large disparity values and occluded regions in stereo image pairs, and further leading to a large number of mismatched points in the obtained disparity map. In response to these challenges, this paper proposes a method to refine the disparity in urban scenes based on open-source GIS data. First, we register the GIS data with the epipolar-rectified images since there always exists unignorable geolocation errors between them. Specifically, buildings with different heights present different offsets in GIS data registering; thus, we perform multi-modal matching for each building and merge them into the final building mask. Subsequently, a two-layer optimization process is applied to the initial disparity map based on the building mask, encompassing both global and local optimization. Finally, we perform a post-correction on the building facades to obtain the final refined disparity map that can be employed for high-precision 3D reconstruction. Experimental results on SuperView-1, GaoFen-7, and GeoEye satellite images show that the proposed method has the ability to correct the occluded and mismatched areas in the initial disparity map generated by both hand-crafted and deep-learning stereo matching methods. The DSM generated by the refined disparity reduces the average height error from 2.2 m to 1.6 m, which demonstrates superior performance compared with other disparity refinement methods. Furthermore, the proposed method is able to improve the integrity of the target structure and present steeper building facades and complete roofs, which are conducive to subsequent 3D model generation.

Despite the rapid development of stereo matching techniques, digital surface models (DSMs) reconstructed from disparities suffer from occlusion, solar shadow, and textureless ambiguity, thus failing to meet the requirements of 3D city modeling. To address this challenge, we proposed a DSM refinement network (RDSMNet) to improve the accuracy and completeness of photogrammetric DSMs. RDSMNet uptakes high-level semantic information from ortho-rectified stereo pairs using Transformer-based UNet architecture to preserve object structures of varying sizes. Meanwhile, a residual branch corrects the DSM values, which is combined with the semantic branch to generate the final refined result. Considering the significant diversity present in remote sensing images, we adopt the mixed-style technique to enhance the generalization capability of DSM refinement. To validate the performance of RDSMNet, we constructed a cross-city, multi-sensor, and multi-resolution dataset. Experimental results demonstrate that the proposed RDSMNet effectively enhances DSM elevation accuracy and structural integrity, exhibiting robust generalization performance across remote sensing data from various source domains. Compared with the current state-of-the-art method, the MAE of the proposed RDSMNet is reduced by 8.64%, and the annotation efficiency is improved from hour-level to minute-level relative to manual editing and involves no human intervention. Code and dataset will be available.

Generating high-precision normalized digital surface models (nDSMs) from a single remote sensing image remains a challenging and ill-posed problem due to the absence of reliable geometric constraints. In this work, we show that monocular depth provides structurally stable cues of local geometry but lacks the global scale and vertical reference required for absolute height recovery. This intrinsic mismatch limits direct depth-to-height regression, particularly when transferring across heterogeneous terrains, land-cover compositions, and imaging conditions. Building on this idea, we propose the Relative Depth–Absolute Height Prediction Network (RDAH-Net), a framework that exploits relative depth as a geometry-aware prior while learning terrain-dependent height mappings from image appearance to absolute height. As the backbone, we employ a lightweight MobileNetV2 enhanced with a Convolutional Block Attention Module (CBAM), and further incorporate a cross-modal bidirectional attention fusion scheme with positional encoding to achieve a deep and effective fusion of image appearance and depth prior cues. Finally, a PixelShuffle-based upsampling strategy is used to sharpen prediction details and mitigate typical upsampling artifacts. Extensive experiments across diverse regions demonstrate that RDAH-Net achieves robust and generalizable height estimation, providing a practical alternative for large-scale mapping and rapid update scenarios.