Junlong Chen

Urban planning and land-use monitoring require not only semantic land-cover maps, but also image-like overhead observations that preserve the spatial organization of urban scenes. Conventional remote-sensing segmentation and image-generation methods usually operate by obtaining available overhead imagery or in reliable overhead conditions. Street-level panoramas are easier to collect and update, but they observe the target scene from a limited terrestrial viewpoint. Recovering overhead information from a single panorama, therefore, requires inferring from an unobserved land-cover footprint before synthesizing a nadir-view remote-sensing image. We propose SDU-G2A, a two-stage framework for aerial image synthesis from a single ground panorama. Stage I predicts a probabilistic BEV land-cover footprint from the input panorama. The proposed Azimuthal-Radial Mapper propagates azimuthal evidence and applies distance-aware supervision to improve semantic recovery in occluded and far-field regions. Stage II converts the inferred footprint into a remote-sensing-like aerial image through diffusion-based synthesis, making the reconstructed urban structure available in an overhead-image form for visual interpretation and geospatial-correspondence evaluation. The main difficulty is that the inferred BEV footprint exhibits spatially varying reliability. Treating it as a deterministic condition can force cross-view segmentation errors into the generated image. SDU-G2A, therefore, derives a pixel-wise uncertainty map from Stage I predictions and uses it as a spatial reliability layer. This design preserves structures with high prediction confidence while allowing uncertain regions to be completed under text-guided and prior-driven generation. Experiments on VIGORv2 show that SDU-G2A improves BEV land-cover estimation, reduces radial prediction errors, and achieves stronger geospatial correspondence and aerial-feature distributional realism than representative G2A baselines. These results indicate that self-derived land-cover uncertainty can bridge cross-view semantic footprint inference and reliable remote-sensing image synthesis from accessible street-level observations.

UAV moving object detection focuses on identifying moving objects in images captured by UAVs, with broad applications in regional surveillance and event reconnaissance. Compared to general moving object detection scenarios, UAV videos exhibit unique characteristics, including foreground sparsity and varying target scales. The direct application of conventional background modeling or motion segmentation methods from general settings may yield suboptimal performance in UAV contexts. This paper introduces an unsupervised UAV moving object detection network. Domain-specific knowledge, including spatiotemporal consistency and foreground sparsity, is integrated into the loss function to mitigate false positives caused by motion parallax and platform movement. Multi-scale features are fully utilized to address the variability in target sizes. Furthermore, we have collected a UAV moving object detection dataset from various typical scenarios, providing a benchmark for this task. Extensive experiments conducted on both our dataset and existing benchmarks demonstrate the superiority of the proposed algorithm.