Few-shot object detection (FSOD) addresses the challenge of limited labeled data by enabling detectors to learn from minimal annotations. Recent work on image–text fusion has shown promise in overcoming data scarcity issues. However, these models suffer from catastrophic forgetting due to two distinct challenges during few-shot adaptation. First, feature space distortion disrupts established relationships between image–text modalities and blurs decision boundaries, leading to inaccuracies in object classification. Second, attention drift impairs the precise visual–textual alignments learned during base training, limiting the model’s ability to focus on relevant regions for accurate object localization. To overcome these limitations, we propose Protoalign, a novel prototype-guided multilayer alignment network that maintains robust cross-modal relationships across multiple network layers while mitigating catastrophic forgetting. Specifically, cross-modal prototype guidance (CPG) enables stable feature learning through class-specific prototype fusion to mitigate feature space distortion. Multimodal feature aggregation (MFA) strengthens feature relationships through channel-level interactions to overcome attention drift. Moreover, we propose an integrator that facilitates consistent information flow between network layers. Protoalign progressively refines multimodal features, enabling effective novel class adaptation while preserving crucial base knowledge. The refined representations are then processed by a detection transformer (DETR) decoder for accurate object detection. Extensive experiments on iSAID, DIOR, FAIR1M-Airplane, and NWPU VHR-10 datasets demonstrate that Protoalign achieves superior performance while significantly reducing catastrophic forgetting compared to existing methods.

Despite DETR-like methods having improved end-to-end detection capabilities, they fundamentally struggle with the uniform processing of entire flattened feature maps, causing queries to attend to irrelevant regions. This results in redundant attention patterns, leading to computational burdens and inefficiencies in the detection of tiny objects in remote sensing imagery across dense and sparse settings. To address these limitations, we present a lightweight Transformer-based encoder-decoder architecture called Dynamic Adaptive Region Transformer (DART). Specifically, Density Adaptive Transformer (DAT) employs an adaptive-region attention (ARA) mechanism that dynamically generates content-aware spatial regions rooted in feature density and semantic relevance. This strategy prioritizes computational resources on semantically rich areas while minimizing focus on irrelevant background regions. Region Aware Decoder (RAD) incorporates a masked region-aware cross-attention (MRA) mechanism, where queries interact exclusively with the adaptive masked regions generated by the encoder, thereby reducing redundant focus on overlapping or irrelevant areas. Meanwhile, a query diversity loss is introduced to penalize overlapping attention patterns among queries, encouraging each query to focus on distinct and complementary regions. By adapting to data density and directing queries to essential areas within the image, DART enhances feature extraction and object localization for various-sized objects in dense and sparse settings. Experimental results demonstrate that DART achieves state-of-the-art performance on AI-TOD, DOTA-v2.0 and LEVIR-Ship benchmarks and exhibits strong generalization capabilities on DIOR, while using only 13 million parameters and a computational cost of 68 GFLOPs.

Existing infrared small target detection algorithms often lack adaptability in complex scenes and heavily rely on parameter configurations. To address this limitation, we propose a novel infrared small target detection method based on adaptive region growing algorithm with iterative threshold analysis that leverages the homogenous compactness of the small target and discontinuity with its surroundings. Initially, the image undergoes adaptive splitting into multiple regions using an automatic seeded region growing (ASRG) algorithm, eliminating the need for preassigned seed points. Next, the segmentation results at each threshold are utilized to calculate the relative residual map (RRM) and local dissimilarity map (LDM), contributing to the selection of the optimal threshold. Finally, RRM and LDM corresponding to the optimal threshold are integrated to accurately characterize the small target signal while effectively removing background clutter. Experimental results show that the proposed method is effective in clutter removal and small target detection in diverse complex scenes, and is robust to the shape and size of targets.

This letter proposes an efficient infrared small target detection method based on the 1-D difference of guided filtering (DoGF). First, the 1-D DoGF is constructed by measuring the difference of image structure fidelity between primitive guided filtering (GF) and local variance weighted GF from the perspective of 1-D signal analysis, which can effectively filter out 1-D noise components and protrude pulse signals. Second, the 1-D row and column DoGF are applied to process the infrared image along horizontal and vertical directions, respectively, and then the row–column crossed DoGF (rcDoGF) and column–row crossed DoGF (crDoGF) are calculated and integrated, which can greatly highlight the pulse-like small target signal and eliminate the background clutter. Finally, the small targets can be extracted with a simple adaptive threshold. Experimental results show that the proposed algorithm has high detection accuracy for small infrared targets under heavy noise interference, as well as for targets with different sizes and shapes.

This letter proposes a novel method based on multiscale dual morphological reconstruction (MDMR) to detect low-contrast ship target with unknown size and polarity (bright or dark) in infrared image. First, the multiscale morphological reconstruction perceives the structural information of various scales. The target signal map (TSM) is presented to indicate the target polarity and enhance target by background subtraction and logarithmic histogram transform. Next, the boundary constraint entropy threshold selection (BCETS) is designed to extract candidate targets and avoid the over- or under-segmentation problems. Then, the local contour contrast descriptor (LCCD) is constructed and the discriminant rule of low-contrast target is established to identify true target and eliminate false alarms. Finally, the pixel-OR operation fuses the multiscale detection results. Extensive experiments show that the proposed method has a better effectiveness and robustness against compared methods.

In the intricate infrared cloudy-sky background, the edge of cloud might be falsely detected as a target because it is usually similar to a small target in contrast and complexity, which are the criteria that conventional small target detection methods adopt to differentiate between background and target. However, the shape of a small target is isotropic and similar to 2-D Gaussian function, while the strong background edge is anisotropic and tends to spread in a certain direction. In this letter, we propose a method to detect small targets in the intricate infrared cloudy-sky background by a local difference adaptive measure (LDAM). This proposed method uses the local structure tensor to perceive the dominant direction and its uncertainty in the local infrared image and then sets the direction and shape of the filter to calculate the local difference. In this way, the background edge is estimated accurately and suppressed effectively. Extensive experiments show that the proposed method outperforms the baseline methods.