Registration for multisensor or multimodal image pairs with a large degree of distortions is fundamental task many remote sensing applications. To achieve accurate and low-cost registration, we propose multiscale framework unsupervised learning, named MU-Net. Without costly ground truth labels, MU-Net directly learns the end-to-end mapping from to their transformation parameters. stacks several deep neural network (DNN) models on multiple scales generate coarse-to-fine registration pipeline,...

Due to the complementary nature of optical and SAR images, their alignment is increasing interest. However, due significant radiometric differences between them, precise matching becomes a very challenging problem. Although current advanced structural features deep learning-based methods have proposed feasible solutions, there still much potential for improvement. In this paper, we propose hybrid method using attention-enhanced (namely AESF), which combines advantages both handcrafted-based...

Although image matching techniques have been developed in the last decades, automatic optical-to-synthetic aperture radar (SAR) is still a challenging task due to significant nonlinear intensity differences between such images. This letter addresses this problem by proposing novel similarity metric for using shape properties. A descriptor named dense local self-similarity (DLSS) first based on self-similarities within Then (named DLSC) defined normalized cross correlation (NCC) of DLSS...

Most traditional approaches classify hyperspectral image (HSI) pixels relying only on the spectral values of input channels. However, spatial context around a pixel is also very important and can enhance classification performance. In order to effectively exploit fuse both structure, we propose novel two-stream deep architecture for HSI classification. The proposed method consists fusion scheme. architecture, one stream employs stacked denoising autoencoder encode each pixel, other takes as...

Image matching is a key preprocessing step for the integrated application of synthetic aperture radar (SAR) and optical images. Due to significant nonlinear intensity differences between such images, automatic them still quite challenging. Recently, structure features have been effectively applied SAR-to-optical image because their robustness differences. However, designed by handcraft are limited achieve further improvement. Accordingly, this letter employs deep learning technique refine...

For remote sensing object detection, fusing the optimal feature information automatically and overcoming sensitivity to adapt multi-scale objects remains a significant challenge for existing convolutional neural networks. Given this, we develop network model with an adaptive attention fusion mechanism (AAFM). The is proposed based on backbone of EfficientDet. Firstly, according characteristics distribution in datasets, stitcher applied make one image containing various scales. Such process...

Over the past few decades, with rapid development of global aerospace and aerial remote sensing technology, types sensors have evolved from traditional monomodal (e.g., optical sensors) to new generation multimodal multispectral, hyperspectral, light detection ranging (LiDAR), synthetic aperture radar (SAR) sensors). These advanced devices can dynamically provide various abundant images (MRSIs) different spatial, temporal, spectral resolutions according application requirements. Since then,...

The supervised deep networks have shown great potential in improving the classification performance. However, training these is very challenging for hyperspectral image given fact that usually only a small amount of labeled samples are available. In order to overcome this problem and enhance discriminative ability network, paper, we propose network architecture super-resolution (SR)-aided with classwise loss (SRCL). First, three-layer SR convolutional neural (SRCNN) employed reconstruct...

We study the problem of object detection in remote sensing images. As a simple but effective feature extractor, Feature Pyramid Network (FPN) has been widely used several generic vision tasks. However, it still faces some challenges when for detection, as objects images usually exhibit variable shapes, orientations, and sizes. To this end, we propose dedicated detector based on FPN architecture to achieve accurate Specifically, considering shapes orientations objects, first replace original...

Remote sensing (RS) image classification has attracted much attention recently and is widely used in various fields. Different to natural images, the RS scenes consist of complex backgrounds stochastically arranged objects, thus making it difficult for networks focus on target objects scene. However, conventional methods do not have any special treatment remote images. In this paper, we propose a two-stream swin transformer network (TSTNet) address these issues. TSTNet consists two streams...

In this letter, a novel object detection method based on feature pyramid network (FPN) is proposed to improve the performance of remote sensing objects. First, since information in background regions may interfere with detection, multi-scale deformable attention module (MSDAM) designed and added top backbone FPN make suppress features while highlight target features. The MSDAM generates maps from receptive fields, thus can fit objects various shapes sizes better predict more precise for...

Automatic registration of synthetic aperture radar (SAR) and optical images is still a challenging problem because the potential differences in geometric intensity. In this work, we propose robust efficient method for improving performance SAR images. Our work consists mainly two steps, including feature point detection stage description stage. first stage, present new extraction method, named nonlinear diffusion-based Harris-Laplace (NDHL) detector, which incorporates diffusion spatial...

Since Hyperspectral Images (HSIs) contain plenty of ground object information, they are widely used in fine-grain classification objects. However, some objects similar and the number spectral bands is far higher than categories. Therefore, it hard to deeply explore spatial–spectral joint features with greater discrimination. To mine HSIs, a Shallow-to-Deep Feature Enhancement (SDFE) model three modules based on Convolutional Neural Networks (CNNs) Vision-Transformer (ViT) proposed. Firstly,...

Automatic matching of multi-modal remote sensing images (e.g., optical, LiDAR, SAR and maps) remains a challenging task in image analysis due to significant non-linear radiometric differences between these images. This paper addresses this problem proposes novel similarity metric for using geometric structural properties We first extend the phase congruency model with illumination contrast invariance, then use extended build dense descriptor called Histogram Orientated Phase Congruency...

Co-registering the Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 optical data of European Space Agency (ESA) is great importance for many remote sensing applications. However, we find that there are evident misregistration shifts between SAR images directly downloaded from official website. To address that, this paper presents a fast effective registration method two types images. In proposed method, block-based scheme first designed to extract evenly distributed interest points....

The structural features using self-similarity have become more popular for multimodal remote sensing image matching. However, mostly because of significant geometric distortions and nonlinear intensity differences between images, these methods produce a limited matching performance when directly applied to images. To address that, we propose novel feature descriptor named pyramid orientated (POSS) matching, which integrates phase congruency (PC) into the model better encoding information....

Due to the rapid development of deep learning techniques and collection large-scale remote sensing datasets, convolutional neural networks (CNNs) have made significant progress in object detection. However, due diversity objects images, multiscale detection is still a challenging task. In this letter, novel framework based on feature pyramid network (FPN) proposed improve performance objects. First, receptive field expansion block (RFEB) designed added top backbone expand FPN adaptively....

Recently, deep learning algorithms, especially feature pyramid network (FPN), have achieved significant progress in object detection of natural scene images. However, due to the complex scenes remote sensing images and diversity objects, FPN still faces following drawback when applied detection. Specifically, original FPN, features each proposal are extracted by RoIAlign. these limited effective receptive fields, making lack crucial contextual information accurately classify locate as well...

Recently, state-of-the-art classification performance of natural images has been obtained by self-supervised learning (S2L) as it can generate latent features through between different views the same images. However, semantic information similar hardly exploited these S2L-based methods. Consequently, to explore potential S2L samples in hyperspectral image (HSIC), we propose nearest neighboring (N2SSL) method, interacting augmentations reliable pairs (RN2Ps) HSI framework bootstrap your own...

Convolutional neural networks (CNNs) have been widely used in remote sensing scene classification. However, the long-range dependencies of local features cannot be taken into account by CNNs. By contrast, a visual transformer (ViT) is good at capturing as it considers global relationship introducing self-attention mechanism. Although ViT can obtain result when training on large-scale datasets, e.g., ImageNet, hard to adapted small-scale datasets (e.g., image datasets). This attributed fact...

Keypoint detection is a crucial step for feature-based image registration. The traditional detectors only extract one type of keypoint such as corner or blob, which not quite beneficial to Accordingly, this letter presents novel detector that aims simultaneously corners and blobs. proposed named Harris-Difference Gaussian (DoG), combines the advantages Harris-Laplace DoG blob detector. In definition Harris-DoG, we first build an scale space by using multiscale Harris Then, these are screened...

Remote sensing image matching is the basis upon which to obtain integrated observations and complementary information representation of same scene from multiple source sensors, a prerequisite for remote tasks such as fusion change detection. However, intricate geometric radiometric differences between multimodal images render registration quite challenging. Although methods have been developed in recent decades, most classical deep learning based techniques cannot effectively extract high...