Infrared small target detection (ISTD) boasts extensive applications across civil and military domains, owing to its exceptional all-day performance. Neural network innovations have led to deep ISTD models that achieve heightened accuracy through extensive datasets. However, these general networks often fail to perceive the sensitivity of small targets and adopt heavy constructions to preserve potential target features, neglecting domain-specific insights and suffering from poor explainability. Our work seeks to rectify this by revisiting the saliency principles inherent to ISTD and developing a learnable local saliency kernel network (L2SKNet). This approach implements a learnable local saliency kernel module (LLSKM) that embodies the concept of “Center subtracts Neighbors,” guiding the network to capture the saliency features (points or edges). We enhance LLSKM by incorporating strategic dilation and structuring it hierarchically, which boosts its capability to capture multiscale infrared features while avoiding parameter explosion. In pursuit of efficiency, we also refine LLSKM into a more compact form by factorizing it into two orthogonal 1-D kernels, yielding a lightweight version. Heatmap visualizations and rigorous quantitative analyses corroborate the effectiveness of our local saliency-guided networks. Comprehensive testing reveals that L2SKNet variants outperform established baselines, demonstrating significant improvements in both visual and numerical assessments. The code is available athttps://github.com/fengyiwu98/L2SKNet.

Infrared small target detection plays a vital role in remote sensing, industrial monitoring, and various civilian applications. Despite recent progress powered by deep learning, many end-to-end convolutional models tend to pursue performance by stacking increasingly complex architectures, often at the expense of interpretability, parameter efficiency, and generalization. These models typically overlook the intrinsic sparsity prior of infrared small targets–an essential cue that can be explicitly modeled for both performance and efficiency gains. To address this, we revisit the model-based paradigm of Robust Principal Component Analysis (RPCA) and propose Dynamic RPCA Network (DRPCA-Net), a novel deep unfolding network that integrates the sparsity-aware prior into a learnable architecture. Unlike conventional deep unfolding methods that rely on static, globally learned parameters, DRPCA-Net introduces a dynamic unfolding mechanism via a lightweight hypernetwork. This design enables the model to adaptively generate iteration-wise parameters conditioned on the input scene, thereby enhancing its robustness and generalization across diverse backgrounds. Furthermore, we design a Dynamic Residual Group (DRG) module to better capture contextual variations within the background, leading to more accurate low-rank estimation and improved separation of small targets. Extensive experiments on multiple public infrared datasets demonstrate that DRPCA-Net significantly outperforms existing state-of-the-art methods in detection accuracy. Code is available at https://github.com/GrokCV/DRPCA-Net.

Deep learning (DL) networks have achieved remarkable performance in infrared small target detection (ISTD). However, these structures exhibit a deficiency in interpretability and are widely regarded as black boxes, as they disregard domain knowledge in ISTD. To alleviate this issue, this work proposes an interpretable deep network for detecting infrared dim targets, dubbed RPCANet. Specifically, our approach formulates the ISTD task as sparse target extraction, low-rank background estimation, and image reconstruction in a relaxed Robust Principle Component Analysis (RPCA) model. By unfolding the iterative optimization updating steps into a deep-learning framework, time-consuming and complex matrix calculations are replaced by theory-guided neural networks. RPCANet detects targets with clear interpretability and preserves the intrinsic image feature, instead of directly transforming the detection task into a matrix decomposition problem. Extensive experiments substantiate the effectiveness of our deep unfolding framework and demonstrate its trustworthy results, surpassing baseline methods in both qualitative and quantitative evaluations. Our source code is available at https://github.com/fengyiwu98/RPCANet.

High-resolution remote sensing image change detection focuses on ground surface changes. It has wide applications, including territorial spatial planning, urban region detection, and military operations. However, class imbalance and pseudo-changes are caused by the unchanged areas far outnumbering the changed areas and lighting changes. To address these problems, we propose spatial-temporal attention with a difference enhancement-based network (STADE-CDNet). In STADE-CDNet, a change detection difference enhancement module (CDDM) is proposed to extract important features from the difference map to detect changed regions. This module enhances the network with differential feature attributes through the training layer, improving the network’s learning ability and reducing the imbalance problem. A temporal memory module (TMM) is designed to extract temporal and spatial information. Inspired by the self-attention mechanism of the transformer, we propose a transformer and TMM (TTMM). Four encoding layers are designed to detect the semantic information from high to low levels of the multitemporal image pairs. The fusion and parallelism of multivariate data are achieved through collaborative modeling of deep learning and change detection, compensating for the need for excessive human intervention in traditional algorithms. We evaluate our approach in two different datasets (LEVIR-CD and DSIFN-CD). Promising quantitative and qualitative results show that STADE-CDNet can improve accuracy. In particular, the proposed CDDM significantly reduces false positive detection, with F1 scores at least 1.97% and 2.1% higher than other methods in the case of the LEVIR-CD and DSIFN-CD datasets, respectively. Our code is available at https://github.com/LiLisaZhi/STADE-CDNet.

Infrared small target detection (ISTD) technique is widely used in infrared searching and tracking (IRST) and military surveillance. Existing detection methods must sufficiently address the challenges of the heterogeneous background with high concealment targets. In this letter, we propose a novel spatial-temporal tensor ring norm regularization (STT-TRNR) to detect infrared small targets. First, to utilize the spatial and temporal context information in a sequence, nonrepetitive spatial-temporal patches are formed by sliding windows in the consecutive frames. The patches are stacked into a tensor structure with spatial and temporal information. Second, the tensor ring nuclear norm is introduced to approximate the rank of the background tensor. The tensor ring regularization improves the correlation between dimensions, protects the internal structure of the tensor, and avoids the dimension disaster caused by train decomposition. Third, the local contrastive feature is used as a priori information to suppress the false alarms caused by the corner edges and other noises and avoid the distortion caused by target movement. Finally, the alternating direction multiplier method (ADMM) is employed to reconstruct the sequence images and retrieve the targets. The experimental results reveal that the suggested model provides improved detection performance and higher robustness across various complicated scene types.

Infrared small target detection (ISTD) is vital for civil and military applications. However, existing methods often face challenges in coping with complex scenes, discriminating targets from similar objects, or leveraging temporal information effectively. To tackle these limitations, we offer an innovative approach that exploits the spatio-temporal structure of infrared images. A four-dimensional (4D) infrared tensor is initially constructed from a sequence of infrared images, and decomposed into lower-dimensional tensors using the tensor train (TT) and its extension – tensor ring (TR) techniques. The ISTD problem is then formulated as a sparse plus low-rank decomposition problem, where the sparse part is the target and the low-rank part is the background. We factorize the composed tensors into matrices via TT and TR unfolding approaches, which mitigates the imbalance between different modes containing spatial and temporal information. By constraining the balanced unfolded components with the weighted sum of nuclear norm, we solve the problem using the alternating direction multiplier method (ADMM). Furthermore, we validate models on several datasets and benchmark them with state-of-the-art techniques in detection accuracy and background suppression. Comparison results demonstrate the superiority of our approach over the existing methods. Moreover, the results of an ablation study with three-dimensional (3D) tensor structures show the effectiveness and feasibility of the dimension expansion to 4D.