The registration of synthetic aperture radar (SAR) and optical images is a challenging task due to the potential nonlinear intensity differences between two images. In this paper, novel image method, which combines diffusion phase congruency structural descriptor (PCSD), proposed for SAR First, reduce influence speckle noise on feature extraction, uniform diffusion-based Harris (UND-Harris) extraction method designed. UND-Harris detector developed based diffusion, proportion, block strategy,...

High-resolution (HR) synthetic aperture radar (SAR) image classification is a challenging task for the limitation of its complex semantic scenes and coherent speckles. Convolutional neural networks (CNNs) have been proven superior local spatial features representation capability SAR images. However, it hard to capture global information images by convolutions. To solve such issues, this letter proposes an end-to-end network named global–local structure (GLNS) HR classification. In GLNS...

With the rapid development of deep learning, significant progress has been made in remote sensing image target detection. However, methods based on learning are confronted with several challenges: (1) inherent limitations activation functions and downsampling operations convolutional networks lead to frequency deviations loss local detail information, affecting fine-grained object recognition; (2) class imbalance long-tail distributions further degrade performance minority categories; (3)...

Although complex-valued (CV) neural networks have shown better classification results compared to their real-valued (RV) counterparts for polarimetric synthetic aperture radar (PolSAR) classification, the extension of pixel-level RV complex domain has not yet thoroughly examined. This paper presents a novel deep fully convolutional network (CV-FCN) designed PolSAR image classification. Specifically, CV-FCN uses CV data that includes phase information and FCN architecture performs labeling....

Multifrequency (MF) polarization synthetic aperture radar (PolSAR) systems can obtain more abundant and continuous earth resource information than single-frequency ones have been widely used in the remote sensing community. However, there are relatively a few researches for fine MF PolSAR image classification, which is an important part of interpretation. The main focus currently on part. Therefore, dual-frequency this article proposes attention fusion network (DFAF-Net). It based...

In this work, we propose a voxel-based single-stage fine-grained and efficient point cloud 3D object detection algorithm to address the inadequate granularity in feature extraction tasks imbalance between efficiency accuracy scenarios. We develop lightweight multibranch cross-sparse convolution network (LMCCN) that is designed preserve of original while achieving enhanced efficiency. Additionally, introduce compact self-attention augmented bird's eye view (BEV) module (CFSAM). This aims...

In recent decades, few-shot object detection in SAR imagery has gained prominence as a major research focus. The unique imaging mechanism of causes the model to suffer from foreground–background imbalance and inaccurate extraction class prototypes for novel instances. Therefore, we propose an innovative algorithm images via context-aware robust Gaussian flow representation. First, design Context-Aware Enhancement module address foreground–context problem by refining representative support...

Rapid advancements in remote sensing (RS) imaging technology have heightened the demand for precise and efficient interpretation of large-scale, high-resolution RS images. Although segmentation algorithms based on convolutional neural networks (CNNs) or Transformers achieved significant performance improvements, trade-off between precision computational complexity remains a key limitation practical applications. Therefore, this paper proposes CVMH-UNet—a hybrid semantic network that...

Compared with the rapid development of single-frequency multi-polarization SAR image classification technology, there is less research on land cover multifrequency polarimetric (MF-PolSAR) images. In addition, current deep learning methods for MF-PolSAR are mainly based convolutional neural networks (CNNs), only local spatiality considered but nonlocal relationship ignored. Therefore, semantic interaction and topological structure, this paper proposes MF semantics topology fusion network...

A pixel-wise classification for high-resolution (HR) synthetic aperture radar (SAR) images is a challenging task, due to the limited availability of labeled SAR data, as well difficulty exploring context information affected by coherent speckle. In this article, we propose novel supervised method HR images, which combines context-aware encoder network (CAEN) and hybrid conditional random field (HCRF) model. First, new CAEN architecture developed based on intrinsic property labeling. The...

The classification of high-resolution (HR) synthetic aperture radar (SAR) images is great importance for SAR scene interpretation and application. However, the presence intricate spatial structural patterns complex statistical nature makes image a challenging task, especially in case limited labeled data. This paper proposes novel HR method, using multi-scale deep feature fusion network covariance pooling manifold (MFFN-CPMN). MFFN-CPMN combines advantages local features global properties...

Synthetic aperture radar (SAR) target detectors based on deep learning have difficulty finding a good balance between accuracy and speed. Current pruning methods are usually used for backbone consistent seldom directly the whole structure of detectors; therefore, edge-end applications, this article proposes new full-link general automatic algorithm SAR detectors, referred to as SARGap. First, SARGap automatically analyzes network by creating dependency graph, divides pair-coupled into same...

Most of the existing deep learning-based SAR target detection algorithms rely on manual experience to repeatedly adjust structures and parameters design models suitable for specific scenarios or tasks. The implementation above methods is complicated, efficiency low, it difficult ensure balance between accuracy complexity. We innovatively propose a hardware-aware algorithm via multiobjective neural architecture search (NAS), referred as SARNas. First, we flexible efficient space, supernet...

The presence of speckles and the absence discriminative features make it difficult for pixel-level polarimetric synthetic aperture radar (PolSAR) image classification to achieve more accurate coherent interpretation results, especially in case limited available training samples. To this end, paper presents a composite kernel-based elastic net classifier (CK-ENC) better PolSAR classification. First, based on superpixel segmentation different scales, three types are extracted consider...

Accurate recognition of the types mainlobe active deception jamming is essential for radar systems to take anti-jamming countermeasures. A bunch deep learning (DL)-based methods that require largescale datasets training have shown promising results. However, capturing a sufficient number diverse samples particularly intricate in actual dynamic and complex battlefields, yielding limited or unbalanced presents significant challenge generalizing DL models. This letter proposes generative model,...

Abstract Active deception jamming recognition has gained significant attention as a crucial aspect of modern electronic warfare, and large quantity methods based on either artificial or deep learning have been proposed to date. In actual complex battlefields, the abundant deceptive signals are extremely difficult obtain leveraged by adversarial jammers often significantly influenced noise. To address these challenges Siamese squeeze wavelet network (SSWAN) for radar active method is...

Convolutional neural networks (CNNs) have been applied to learn spatial features for high-resolution (HR) synthetic aperture radar (SAR) image classification. However, there has little work on integrating the unique statistical distributions of SAR images which can reveal physical properties terrain objects, into CNNs in a supervised feature learning framework. To address this problem, novel end-to-end classification method is proposed HR by considering both context and features. First,...

Pedestrian trajectory prediction is extremely challenging due to the complex social attributes of pedestrians. Introducing latent vectors model multimodality has become latest mainstream solution idea. However, previous approaches have overlooked effects redundancy that arise from introduction vectors. Additionally, they often fail consider inherent interference pedestrians with no history during training. This results in model’s inability fully utilize training data. Therefore, we propose a...