Xiaolin Tian

One-bit quantization technology effectively simplifies the hardware architecture of synthetic aperture radar (SAR) and significantly reduces the volume of echo data, making it a pivotal research direction in spaceborne SAR systems. However, the high computational cost and latency associated with one-bit SAR signal reconstruction have limited the efficiency of downstream target detection applications. To address the demand for efficient ship target detection in one-bit SAR imagery, a label self-supervised ship detection training method on one-bit non-reconstructed SAR images is proposed in this paper, which enables generic deep learning models to perform detection directly in one-bit SAR images based on the chirp scaling algorithm (CSA) without signal recovery during inference. First, a training framework based on curriculum learning is designed, aiming to discriminate between real ship targets and false targets. Second, a one-bit reconstruction and training label self-supervised correction branch is incorporated into the training framework to autonomously generate false target labels and progressively guide the model to learn false target features. Third, a one-bit non-reconstructed image dataset is proposed to validate the trained model, which maps only reliable ship target annotations from full-precision echo imaging results. Finally, experiments conducted on real SAR data demonstrate that the detection model derived from the proposed training method can directly detect ship targets in the non-reconstructed images. Compared with the classic “reconstruction-and-detection” workflow, the trained model runs inference, bypassing the reconstruction process, and achieves a 20-fold improvement in efficiency without compromising detection accuracy. This method provides a practical solution for the rapid acquisition of target information in large-scale ship monitoring using one-bit SAR.

Automated detection of lunar impact craters from digital elevation model (DEM) data is important for lunar geological analysis, landing-site selection, and crater catalog updating. However, this task remains challenging because lunar craters exhibit large scale variations, weak or degraded rims, ambiguous boundaries, and complex topographic backgrounds. In addition, large-scale lunar remote sensing applications require detection models to achieve a reasonable balance among accuracy, model complexity, and inference efficiency. To address these challenges, this study proposes FPW-YOLO11n, a frequency-perception crater detection method developed based on YOLO11n. First, a Frequency-Directional Attention Module (FDA-Module) is introduced into the shallow stage of the backbone. This module combines frequency-aware channel attention and direction-aware spatial attention to enhance the representation of crater rim structures, elevation variations, and directional topographic cues in DEM data. Second, a C2PSA-LRSA module is designed by embedding Local Region Self-Attention into the C2PSA framework, thereby improving local contextual feature interaction while reducing the excessive cost associated with global self-attention. Third, Inner-WIoU is adopted to replace the original CIoU loss in YOLO11n. By combining the auxiliary-box mechanism of Inner-IoU with the sample-quality-aware weighting strategy of WIoU, Inner-WIoU provides a more flexible bounding-box regression objective for craters with weak rims, scale variations, and uncertain boundaries. A DEM-based lunar crater dataset was constructed from the Moon LRO LOLA–SELENE Kaguya TC DEM Merge 60N60S 59m product and the Robbins lunar crater catalog, covering the non-polar region from 60° S to 60° N and containing 4760 image tiles. Under the random data-splitting strategy, FPW-YOLO11n achieves 78.3% Precision, 66.2% Recall, 75.1% mAP@0.5, and 50.2% mAP@0.5:0.95, outperforming the YOLO11n baseline by 1.2, 2.0, 1.6, and 4.0 percentage points, respectively. Additional experiments based on geographically disjoint data splitting further show that the proposed method consistently performs better than YOLO11n on DEM data, indicating that the proposed structural improvements remain effective under a more rigorous spatially independent evaluation setting. Although the computational cost increases from 6.3 to 24.0 GFLOPs, FPW-YOLO11n maintains a compact parameter size of 2.59 M and a high inference speed, demonstrating an improved accuracy–efficiency trade-off for lunar crater detection from DEM data.

Unmanned aerial vehicle (UAV) object detection plays an essential role in modern visual perception, but it remains challenging because UAV imagery typically contains extremely small, densely distributed objects embedded in complex backgrounds. Conventional detectors, including the recent YOLOv12, are prone to losing critical spatial details during downsampling and often exhibit task misalignment between classification and localization, particularly under severe scale variations. To address these problems, this study proposes SPTD-YOLO, a small-object-aware pyramidal and task-aligned dynamic detector. Specifically, a Small Object Enhanced Pyramid (SOEP) is developed by incorporating SPDConv and CSPOmniKernel to preserve and refine shallow, fine-grained features. In addition, a high-resolution P₂ detection layer is introduced to increase spatial grid density and strengthen the structural representation of tiny objects. Furthermore, a Task-Aligned Dynamic Detection Head (TADDH) is designed to decouple and coordinate classification and regression through dynamic convolution and a synergistic dual-gating mechanism. Experiments on VisDrone2019 show that SPTD-YOLO improves mAP@0.5 by 8.37% and mAP@0.5:0.95 by 5.11% over YOLOv12 while maintaining practical efficiency for UAV edge deployment.

Satellite-based assessment of power tower damage is essential for rapid disaster response but is challenged by the scarcity of damage samples and the cross-scale mismatch between close-range UAV imagery and satellite imagery. Existing data augmentation methods, including copy-based strategies and diffusion-based generation, often fail to produce reliable samples due to their dependence on the training data distribution and the lack of explicit control over object scale and domain discrepancy. To address these issues, we propose a scale-constrained and frequency-adaptive diffusion-based data construction framework that explicitly models the scale distribution prior of power towers in the remote sensing domain and incorporates frequency-domain adaptation before image generation. Specifically, scale-aware instance embedding is used to construct training samples that conform to satellite-scale statistics, while frequency-domain adaptation is introduced to reduce spectral and texture discrepancies between UAV-derived damaged references and satellite imagery. A diffusion-based inpainting model is then trained on the constructed dataset to reconstruct damage at original tower locations. The experimental results, including feature statistical analysis and downstream change detection validation, demonstrate that the proposed method achieves better alignment with real satellite-scale distributions, reduces geometric and spectral–textural inconsistencies, and improves boundary continuity and structural realism under cross-resolution conditions.

Spaceborne interrupted frequency-modulated continuous-wave synthetic aperture radar (iFMCW SAR) employs a single antenna on a single spacecraft operating in a time-division transmit/receive mode, effectively avoiding mutual interference between transmitted and received signals and thereby overturning the design paradigm of spaceborne FMCW SAR systems. However, the periodic switching of the antenna between transmit and receive states results in periodic data gaps along the azimuth direction in the echo signal, leading to spurious artifacts in the reconstructed images and severely degrading image quality. Sparse signal recovery techniques based on compressive sensing models have been shown to effectively suppress such spurious targets. Nevertheless, the generalized orthogonal matching pursuit (GOMP) algorithm requires prior knowledge of the signal sparsity, a condition that is often impractical in real-world scenarios. To address this limitation, this paper investigates the variation pattern of the residual norm with respect to sparsity in the GOMP algorithm and proposes a modified GOMP algorithm based on binary search. This approach enables rapid and accurate determination of the true sparsity level without prior knowledge, thereby achieving sparsity-adaptive reconstruction with GOMP and significantly enhancing the imaging quality of iFMCW SAR. Simulation experiments involving both point and scene targets are provided to demonstrate the effectiveness and potential of the proposed algorithms for practical applications.

Martian crater detection plays a critical role in landing-site selection and route planning, and it directly influences whether a mission can be executed safely and whether scientific observations and surface material information can be reliably acquired and returned. These outcomes, in turn, affect subsequent investigations of Martian geology and environmental evolution. Despite the progress of recent detectors, missed detections still occur, especially for medium-to-large craters with weak or blurred rims and for small overlapping craters. At the same time, practical deployment often requires a lightweight architecture so that computational cost can be controlled under real operational constraints. To address these issues, we propose SKE-YOLO11, a lightweight model designed for robust crater detection. First, we construct KWConv and C3K2_KW to replace the standard convolution and the C3K2 module in YOLO11n. This design reduces computation while strengthening feature extraction for small craters. Second, we introduce a Strip Convolution Module (SCM) to enlarge the effective receptive field, which helps the network learn rim and texture cues of medium-to-large craters and reduces missed detections. Third, considering the geometric characteristics of crater annotations, we fuse EIoU and Inner-IoU into an Inner-EIoU loss to replace the CIoU used in YOLO11n, thus improving bounding-box regression. Experiments show that SKE-YOLO11 achieves 83.0% Precision, 75.0% Recall, and 82.8% mAP@0.5. Compared with YOLO11n, Recall and mAP@0.5 improve by 2.3% and 1.8%, and parameters and GFLOPs decrease by 0.08 M and 1.3, respectively.

The precise segmentation of geological linear features, spanning from planetary lineaments to terrestrial fractures, demands capturing long-range dependencies across complex anisotropic topologies. Although State Space Models (SSMs) offer near-linear computational complexity, their dependence on rigid, axis-aligned scanning trajectories induces a fundamental topological mismatch with curvilinear targets, resulting in fragmented context and feature erosion. To bridge this gap, we propose Fluxamba, a lightweight architecture that introduces a topology-aware feature rectification framework. Central to our design is the Structural Flux Block (SFB), which orchestrates an anisotropic information flux by integrating an Anisotropic Structural Gate (ASG) with a Prior-Modulated Flow (PMF). This mechanism decouples feature orientation from spatial location, dynamically gating context aggregation along the target's intrinsic geometry rather than rigid paths. Furthermore, to mitigate serialization-induced noise in low-contrast environments, we incorporate a Hierarchical Spatial Regulator (HSR) for multi-scale semantic alignment and a High-Fidelity Focus Unit (HFFU) to explicitly maximize the signal-to-noise ratio of faint features. Extensive experiments on diverse geological benchmarks (LROC-Lineament, LineaMapper, and GeoCrack) demonstrate that Fluxamba establishes a new state-of-the-art. Notably, on the challenging LROC-Lineament dataset, it achieves an F1-score of 89.22% and mIoU of 89.87%. Achieving a real-time inference speed of over 24 FPS with only 3.4M parameters and 6.3G FLOPs, Fluxamba reduces computational costs by up to two orders of magnitude compared to heavy-weight baselines, thereby establishing a new Pareto frontier between segmentation fidelity and onboard deployment feasibility.

Few-shot recognition in synthetic aperture radar (SAR) imagery remains a critical bottleneck for real-world applications due to extreme data scarcity. A promising strategy involves synthesizing a large dataset with a generative adversarial network (GAN), pre-training a model via self-supervised learning (SSL), and then fine-tuning on the few labeled samples. However, this approach faces a fundamental paradox: conventional GANs themselves require abundant data for stable training, contradicting the premise of few-shot learning. To resolve this, we propose the consistency-regularized generative adversarial network (Cr-GAN), a novel framework designed to synthesize diverse, high-fidelity samples even when trained under these severe data limitations. Cr-GAN introduces a dual-branch discriminator that decouples adversarial training from representation learning. This architecture enables a channel-wise feature interpolation strategy to create novel latent features, complemented by a dual-domain cycle consistency mechanism that ensures semantic integrity. Our Cr-GAN framework is adaptable to various GAN architectures, and its synthesized data effectively boosts multiple SSL algorithms. Extensive experiments on the MSTAR and SRSDD datasets validate our approach, with Cr-GAN achieving a highly competitive accuracy of 71.21% and 51.64%, respectively, in the 8-shot setting, significantly outperforming leading baselines, while requiring only ~5 of the parameters of state-of-the-art diffusion models. Code is available at: https://github.com/yikuizhai/Cr-GAN.

Lunar Impact Craters are geologic formations that contain vestiges of the Moon’s historical evolution. Detecting and analysing them at the lunar surface is a crucial issue. This paper proposed an improved YOLO v11 model called YOLO v11_MSEF. The enhanced interlayer feature correlation (EFC) mechanism adds the channel attention model and local attention model to let the YOLO v11 enhance expressive power for multi-scale feature fusion, which is named YOLO v11_MSEF. The datasets are set as 9500 labelled targets for training and 2500 for validation, which were applied in this project. The YOLO v11_MSEF mode achieves a precision performance of 70.40%, which is a significant improvement over the 68.90% achieved by the YOLO v11, the remaining performance parameters also demonstrate comparable or superior results.

Although deep learning has advanced remote sensing change detection (RSCD), most methods rely solely on image modality, limiting feature representation, change pattern modeling, and generalization especially under illumination and noise disturbances. To address this, we propose MMChange, a multimodal RSCD method that combines image and text modalities to enhance accuracy and robustness. An Image Feature Refinement (IFR) module is introduced to highlight key regions and suppress environmental noise. To overcome the semantic limitations of image features, we employ a vision language model (VLM) to generate semantic descriptions of bitemporal images. A Textual Difference Enhancement (TDE) module then captures fine grained semantic shifts, guiding the model toward meaningful changes. To bridge the heterogeneity between modalities, we design an Image Text Feature Fusion (ITFF) module that enables deep cross modal integration. Extensive experiments on LEVIRCD, WHUCD, and SYSUCD demonstrate that MMChange consistently surpasses state of the art methods across multiple metrics, validating its effectiveness for multimodal RSCD. Code is available at: https://github.com/yikuizhai/MMChange.