Jinyong Chen

Ship detection in range-compressed (RC) synthetic aperture radar (SAR) images requires recognizing spatially extended responses rather than compact target appearances. Because elongated responses are not unique to ships, clutter can interfere with both feature aggregation and candidate selection. We propose FM-DETR, a frequency–morphology collaborative Transformer that extends RC-aware feature modeling to decoder-query initialization. An asymmetric cross-scale feature pyramid uses orthogonal strip convolutions to accommodate the unequal range and azimuth extents of target responses. A Fourier-gated intra-scale interaction module complements spatial self-attention with learnable spectral modulation, while a content-dependent gate regulates the spectral residual. The refined representation provides spatial guidance for multiscale candidate scoring and query selection. Evaluation follows the official RCShip-1.0 split, with synthetic training samples and scene-derived test samples reconstructed in the RC domain. FM-DETR achieves average precision (AP@0.5) of 90.14%, a precision of 89.53%, and a recall of 84.42%. With the backbone and training settings held fixed, the complete model improves AP@0.5 by 4.00 percentage points over the baseline. These results support integrating direction-sensitive aggregation, spectral refinement, and guided query initialization for ship detection in RC images.

Hyperspectral imagery enables precise land cover characterization via continuous spectral measurements, but severe inter-band redundancy necessitates unsupervised band selection (UBS) as a critical preprocessing step. Existing subspace partitioning-based UBS methods suffer from high computational complexity, tight coupling between partitioning and selection criteria, and a lack of rigorous physical prior support, while the counterintuitive superior performance of simple uniform sampling still lacks theoretical interpretation. To address these gaps, this paper proposes a universal UBS enhancement framework grounded in the inherent Markov property of hyperspectral bands. We first design the spectral Markov index (SMI) to quantitatively characterize the spectral Markov property, and theoretically prove that continuous uniform partitioning is the globally optimal strategy under this prior. The proposed two-stage framework fully decouples subspace partitioning and selection criteria, enabling non-intrusive performance enhancement for conventional UBS methods. Systematic experiments on five benchmark datasets validate that the proposed framework achieves an 86.67% performance improvement probability, and confirm that uniform spectral coverage is the dominant determinant of UBS performance.

Shore-based maritime surveillance radars suffer from systematic range and azimuth errors that degrade target-tracking accuracy. This paper proposes a Vision Transformer (ViT) variant that corrects these errors using Automatic Identification System (AIS) data as the ground truth, modelling nonlinear error patterns via self-attention without requiring explicit physical models of the underlying error sources. Evaluated on the Maritime Target Detection and Tracking (MTDSP) dataset (≈80,000 paired radar-AIS observations), the proposed method reduces range mean absolute error (MAE) by 98.5% (514.76 m → 7.77 m) and azimuth MAE by 89.8% (1.37° → 0.14°) relative to uncalibrated measurements. Controlled experiments isolating architectural components confirm that self-attention, patch embedding, and multi-task learning each contribute measurable gains, particularly in tail-error robustness. These results demonstrate the viability of Transformer-based architectures for correcting radar systematic errors in maritime surveillance.

There are differences in imaging characteristics between infrared and visible light images: visible light images can provide rich texture and color information, but imaging is limited in harsh weather conditions. Infrared images are based on the target’s thermal radiation characteristics and have the ability to resist environmental interference but lack details and background information. Effectively integrating the two can significantly enhance scene understanding ability and improve environmental perception and target recognition performance in applications such as intelligent driving. However, existing fusion methods still face challenges, especially in complex scenes where it is difficult to balance the full preservation of target information with the complete presentation of background details, often resulting in difficulties in extracting differentiated features from different modalities. This article proposes a target detection method based on the visible light infrared fusion diffusion model. This method introduces the Stable Diffusion architecture and designs a target perception spatial fusion weight module that can adaptively generate a spatial fusion weight map based on modal differences. By implementing a multi-stage dynamic fusion strategy, the fusion ratio is automatically adjusted at different diffusion stages. A full-step multi-step prediction mechanism is adopted to improve fusion quality and stability. Compared with existing methods, the method proposed in this article has significant advantages. Experiments on multiple publicly available datasets have shown that this method outperforms existing mainstream methods in key metrics such as Peak Signal to Noise Ratio (PSNR), Mean Square Error (MSE), and ean Absolute Error (MAE) and also demonstrates good detection performance in downstream tasks for object detection.

Compared with generic object detection, remote sensing image small object detection typically encounters problems such as information loss, noise characteristic interference, and disturbance of the bounding box. Different from the con ventional method of bounding box regression, we propose a self-learning feature point network (SFPNet) for small object detection, which can captures the geometric information of the object in arbitrary directions as well as enhance the representation of small objects by leveraging feature points. Specifically, a feature points orientation transformation module is proposed to accurately obtain classification and localization information. To improve the learning ability of weak sample objects, a positive and negative sample evaluation and balancing strategy is proposed. Additionally, we introduce a self-learning method of feature points under spatial constraints, which enhances the self-refinement and object expression ability of the feature points while eliminating false alarms. To enhances the training of hard-to-distinguish samples, we propose online feature point hard example mining. Experimental results on three benchmark datasets (including AI-TOD, SODA-A, and DOTA v2) show that the proposed method achieves more than 1.3% improvement in small object detection accuracy at different scales.

In the field of low-altitude aerial drone data fusion, the fusion of infrared and visible light images remains challenging due to issues such as large modal differences, insufficient cross-modal alignment, and limited global context modeling. Traditional methods struggle to extract complementary information across modalities, while deep learning methods often lack sufficient global receptive fields (convolutional neural networks) or fail to preserve local details (standard Transformers). To address these issues, we propose a Cross-modal Guided Dual-Branch Network (CGDBN) that combines convolutional neural networks and Transformer architecture. Our framework contribution: We designed a Target-modal Feature Extraction Mechanism (TMFEM) module with specialized thermal characteristics for infrared feature extraction, which does not require processing of visible light features; we introduced Simplified Linear Attention Blocks (SLABs) into our framework to improve global context capture as a module; we designed a Cross-Modal Interaction Mechanism (CMIM) module for bidirectional feature interaction; and we designed a Density Adaptive Multimodal Fusion (DAMF) module that weights modal contributions based on content analysis. This asymmetric design recognizes that different types of images have different characteristics and require targeted processing. The experimental results on AVMS, M3FD, and TNO datasets show that the proposed model has a peak signal-to-noise ratio (PSNR) of 16.2497 on the AVMS dataset, which is 0.9971 higher than the best benchmark method YDTR (peak signal-to-noise ratio: approximately 15.2526). The peak signal-to-noise ratio on the M3FD dataset is 16.5044, which is 0.7480 higher than the best benchmark method YDTR (peak signal-to-noise ratio of approximately 15.7564). The peak signal-to-noise ratio on the TNO dataset is 17.3956, which is 0.7934 higher than the best benchmark method YDTR (peak signal-to-noise ratio: approximately 16.6022), and the overall performance on all other indicators is among the top in all comparison models. This method has broad application prospects in fields such as drone data fusion.

Rapid urbanization drives significant land use transformations, making the timely detection of newly constructed buildings a critical research focus. This study presents a novel unsupervised framework that integrates pixel-level change detection with object-level, mono-temporal building information to identify new constructions. Within this framework, we propose the Building Line Index (BLI) to capture structural characteristics from building edges. The BLI is then combined with spectral, textural, and the Morphological Building Index (MBI) to extract buildings. The fusion weight (φ) between the BLI and MBI was determined through experimental analysis to optimize performance. Experimental results on a case study in Wuhan, China, demonstrate the method’s effectiveness, achieving a pixel accuracy of 0.974, an average category accuracy of 0.836, and an Intersection over Union (IoU) of 0.515 for new buildings. Critically, at the object-level—which better reflects practical utility—the method achieved high precision of 0.942, recall of 0.881, and an F1-score of 0.91. Comparative experiments show that our approach performs favorably against existing unsupervised methods. While the single-case study design suggests the need for further validation across diverse regions, the proposed strategy offers a robust and promising unsupervised pathway for the automatic monitoring of urban expansion.

Synthetic aperture radar imagery has shown a significant potential especially in maritime surveillance, making SAR-based object detection a prominent research area. However, detecting ships in SAR images remains a challenge due to its complex background interference, densely arranged ship formations, and diverse target types. Faced with these problems, this paper proposes SFONet, a joint spatial-frequency (SF) domain algorithm for multiclass ship-oriented detection. A novel SF attention module is introduced to capture and integrate spatial and frequency domain features effectively. Moreover, with the branch processing and convolutional residual enhancement operations, the extracted information is refined deeply and fused appropriately. Consequently, it can mitigate the impact of complex background interference and preserve critical details for accurate object detection. In case of the angle prediction, the traditional oriented bounding box is mapped to a Gaussian bounding box for target encoding. In the meanwhile, a strategy combining probabilistic intersection-over-union and distribution focal loss enhances the prediction accuracy significantly. Extensive experiments based on SSDD+, HRSID, RSDD, and SRSDD datasets are provided, demonstrating the superior performance of SFONet in both single-class and multiclass ships oriented detection tasks. Ablation studies further validate the key components for improving the accuracy of detection and classification.