Pengjie Tao

The rapid proliferation of open-access vector building footprint datasets has created substantial opportunities for large-scale urban analysis, yet their reliability remains difficult to assess effectively and consistently. This challenge arises primarily from the scarcity of large-scale, high-fidelity reference benchmarks across diverse urban environments. To address these limitations, we construct a benchmark-grade reference dataset comprising over 200,000 manually annotated building footprints across 10 purposively selected cities across diverse global regions through an exhaustive curation campaign and strict manual-automated quality validation. Using this dataset, we develop a three-tier evaluation framework that assesses statistical consistency, object recovery and spatial accuracy, and morphology fidelity among matched building pairs. Evaluating 11 mainstream datasets reveals pervasive underestimation in both building area (average bias: −14.9%) and count (−34.0%). Average city-level intersection-over-union values range from 29.1% to 58.3%, indicating low-to-moderate spatial overlap. For matched building pairs, the datasets demonstrate relatively high geometric fidelity, with morphological similarity scores of 0.495 to 0.803. Pronounced regional performance variations indicate that no single dataset is universally optimal. Specifically, Microsoft Global ML Building Footprints offers the most balanced multi-city performance, Google Research Open Buildings is recommended for boundary shape fidelity, and regional datasets remain competitive for local analyses.

High-resolution satellite stereo matching is fundamental for 3D mapping. While current deep learning models achieve impressive macroscopic geometric accuracy, they inherently rely on convolutions with static receptive fields. This rigidity struggles with extreme scale variations across complex landscapes, often causing pervasive localized matching errors, such as edge-fattening artifacts at object boundaries. Furthermore, applying standard dynamic convolutions to address this within massive 4D cost volumes incurs prohibitive memory overhead. To address this trade-off, we propose a parameter-efficient, scale-adaptive cost aggregation strategy. We introduce a Multi-scale Dynamic Receptive Field (MDRF) mechanism that decouples the learned base operator from the dilation-dependent sampling support. By reusing a single shared kernel across multiple dilations via a content-aware attention gate, the network adaptively weights multi-scale context responses. This plug-and-play module is parameter-efficient, with memory and throughput explicitly reported in the evaluated configurations. Evaluated on the WHU-Stereo and US3D benchmarks, our approach consistently lowers the outlier rate, achieving absolute D1-error reductions of 0.43 and 0.49 percentage points, respectively, compared with the baseline. Error-map analysis shows that MDRF reduces localized boundary artifacts, supporting a measured accuracy-efficiency trade-off for practical remote sensing applications.

Advances in unmanned aerial vehicle (UAV) technology have significantly lowered the barriers to achieving close-range photogrammetry through aerial imaging, enabling the widespread application of object-level real-scene 3D modeling in large-scale scenarios. However, existing photogrammetric camera network construction methods for object-level 3D modeling lack explicit constraints on the regularity of UAV shooting points, resulting in frequent start-stop cycles and a reliance on manual operations. This significantly reduces the efficiency of UAV data collection and demonstrates suboptimal performance in balancing modeling accuracy and flight efficiency. To address this, building on the framework of our previously proposed concept of the nap-of-the-object photogrammetry, we propose a scene-adaptive camera network construction method aimed at improving the time and energy efficiency of UAV photography in ultra-fine (i.e., millimeter-level) 3D modeling. Using a pre-acquired coarse point cloud of the target scene, we divide the scene into supervoxels and perform an initial clustering into several regions. Then, we apply a Markov random field (MRF) optimization to refine the point cloud segmentation, approximating each segmented region as a plane while minimizing their number. Finally, strip-wise camera network construction is applied to each segmented region to generate the final camera network. Experimental results on two buildings at Wuhan University demonstrate that, compared to existing methods, our approach achieves high-precision detailed modeling while improving the time and energy efficiency, and also outperforms other relative methods in balancing modeling precision and flight efficiency.

The equator’s unique combination of high humidity and temperature renders optical satellite imagery highly susceptible to persistent cloud cover. In contrast, synthetic aperture radar (SAR) offers a robust alternative due to its ability to penetrate clouds with microwave imaging. This study addresses the challenges of cloud-induced data gaps and cross-sensor geometric biases by proposing an advanced optical and SAR image-matching framework specifically designed for cloud-prone equatorial regions. We use a prompt-driven visual segmentation model with automatic prompt point generation to produce cloud masks that guide cross-modal feature-matching and joint adjustment of optical and SAR data. This process results in a comprehensive digital orthophoto map (DOM) with high geometric consistency, retaining the fine spatial detail of optical data and the all-weather reliability of SAR. We validate our approach across four equatorial regions using five satellite platforms with varying spatial resolutions and revisit intervals. Even in areas with more than 50 percent cloud cover, our method maintains sub-pixel edging accuracy under manual check points and delivers comprehensive DOM products, establishing a reliable foundation for downstream environmental monitoring and ecosystem analysis.

Rapid and reliable assessment of building damage is essential for post‐disaster response and recovery. As windows often reflect critical structural changes, their automatic extraction from street‐view images provides valuable insights for emergency assessment, urban risk modeling, and disaster database updates. Existing methods struggle to leverage the quadrilateral prior of windows due to two main issues: poor handling of perspective distortion and the lack of robust loss functions when precise vector annotations are unavailable. To overcome these challenges, we introduce QuadWindow, a framework specifically designed to handle perspective distortions through a perspective transformation sub‐network that predicts transformations from street‐view images to frontal views, significantly simplifying window extraction tasks without manual correction. Additionally, we propose a differentiable rendering loss that directly aligns predicted quadrangles with raster‐based ground truth, bypassing the need for explicit corner‐point annotations. Experimental results demonstrate that QuadWindow outperforms state‐of‐the‐art methods across five façade datasets, with an average F1‐score of 87.6% and Intersection over Union (IoU) of 78.03%, achieving 1.47% and 5.2% improvement, respectively.

High-precision elevation mapping is essential for ecological restoration, marine disaster assessment, and morphodynamic simulation in intertidal zones. Current methodologies are often impeded by an over-reliance on extensive in-situ measurements and are typically applicable only to regions devoid of vegetation. In this study, we first propose a novel method for spatially continuous elevation mapping of large-scale muddy intertidal zones within highly turbid estuaries, utilizing features at pixel, neighborhood, and temporal scales from satellite multispectral images. This method utilizes a random forest (RF) to model the relationships between elevations from ICESat-2 (Ice, Cloud, and Elevation Satellite 2) and band, texture, and index features from Sentinel-2, without relying on any supplementary in-situ measurements. The innovation and strength of the proposed method lie in the simultaneous incorporation of two temporal features: vegetation occurrence frequency and water inundation frequency. These two features effectively utilize the variations observed in different regions and land covers within the Sentinel-2 image series caused by the unique tide periodic fluctuation phenomenon and elevation trend law in intertidal zones, thereby rendering the method applicable to elevation prediction across the entire spatial range of intertidal zones, rather than being limited to non-vegetated regions. A case study conducted on the muddy intertidal zones of the islands in the Yangtze River Estuary from 2019 to 2023 reveals that the average root mean square errors are 0.33 m and 0.69 m for high-mid and mid-low intertidal zones, respectively. The proposed method demonstrates superior performance in terms of vertical accuracy, spatiotemporal resolution, and spatial continuity in comparison to the state-of-art waterline detection and inundation frequency methods.