Recently, the field of hyperspectral image (HSI) classification is dominated by deep learning-based methods. However, training learning models usually needs a large number labeled samples to optimize thousands parameters. In this article, multiview method proposed deal with small sample problem HSI. First, two views an HSI scene are constructed applying principal component analysis different bands. Second, residual network designed embed latent space. The trained maximizing agreement between...

Extracting buildings from very high-resolution satellite images is a challenging yet important task for applications such as urban monitoring. Multiscale feature learning proves to be potential solution toward accurate extraction of buildings. This study exploits powerful multiscale module, hierarchical vision transformer by shifted windows (swin), backbone within building network. To this end, we first designed general structure extraction, consisting extract features and head network fuse...

We present an efficient multi-view stereo (MVS) network for 3D reconstruction from images. While previous learning based approaches performed quite well, most of them estimate depth maps at a fixed resolution using plane sweep volumes with hypothesis each plane, which requires densely sampled planes desired accuracy and therefore is difficult to achieve high maps. In this paper we introduce coarse-to-fine inference strategy depth. This first estimates the map coarsest level, finer levels are...

Land cover classification (LCC) is essential for monitoring land use and changes. This study examines the integration of optical (OPT) synthetic aperture radar (SAR) images precise LCC. The disparity between OPT SAR introduces challenges in fusing high-level semantic information utilizing multi-scale features. To address these challenges, this paper proposes a novel multi-modal capsules model (M²Caps) incorporating learning cascaded features fusion modules. module models abstract...

Deep learning based methods have made great progress in hyperspectral image classification. However, training a deep model often requires large number of labeled samples, which are not always available practical applications. In this paper, simple but innovative classification paradigm to exploit morphological attribute profile cube is proposed improve the small sample performance image. First, profiles constructed by applying different filters Morphological cubes then extracted as feature...

With geospatial intelligence research developing rapidly, automatic road extraction is becoming a fundamental and challenging task. Due to the special geometric structure spectral information of networks, existing methods suffer from incomplete fractured results. In this work, novel convolutional neural network, incorporating boundary details junction via dual-branch multi-task structure, proposed learn synergistic feature representations strengthen connectivity. Firstly, BiFPN-based...

Extracting buildings from remote sensing images is an important task with a variety of applications. Considerable attention has focused on achieving new SOTA accuracy more and advanced deep learning models. However, the developed models still hardly generalize across geographical areas, hindering practical use approaches. To attack this problem, we established baseline for model cross-area generalization ability using available datasets BE. In addition to two popular FCN-based models, first...

The annotations used during the training process are crucial for inference results of remote sensing images (RSIs) based on a deep learning framework. Unlabeled RSIs can be obtained relatively easily. However, pixel-level annotation is that necessitates high level expertise and experience. Consequently, use small sample methods has attracted widespread attention as they help alleviate reliance large amounts high-quality labeled data current methods. Moreover, research still in its infancy...

Deep learning, which is a dominating technique in artificial intelligence, has completely changed image understanding over the past decade. As consequence, sea ice extraction (SIE) problem reached new era. We present comprehensive review of four important aspects SIE, including algorithms, datasets, applications and future trends. Our focuses on research published from 2016 to present, with specific focus deep-learning-based approaches last five years. divided all related algorithms into...

Recently, the deep learning models have achieved great success in hyperspectral images (HSI) classification. However, most of fail to obtain satisfactory results under condition small samples due contradiction between large parameter space and insufficient labeled HSI. To address problem, a model based on induction network is designed this article improve classification performance HSI samples. Specifically, typical meta-training strategy adopted, enabling acquire stronger generalization...

In the existing unsupervised domain adaptation (UDA) methods for remote sensing images (RSIs) semantic segmentation, class symmetry is a widely followed ideal assumption, where source and target RSIs have exactly same space. practice, however, it often very difficult to find RSI with classes as RSI. More commonly, there are multiple available. And always an intersection or inclusion relationship between spaces of each source–target pair, which can be referred asymmetry. Nevertheless,...

Deep learning has been widely used in hyperspectral image (HSI) classification. However, a deep model is data-driven machine method, and collecting labeled data quite time-consuming for an HSI classification task, which means that needs lot of cannot deal with the small sample problem. We explore problem graph convolutional network (GCN). First, number samples are treated as graph. Then, GCN (an efficient variant neural networks) operates directly on constructed from HSI. utilizes adjacency...

Deep-learning-based methods have seen a massive rise in popularity for hyperspectral image classification over the past few years. However, success of deep learning is attributed greatly to numerous labelled samples. It still very challenging use only samples train models reach high accuracy. An active deep-learning framework trained by an end-to-end manner is, therefore, proposed this paper order minimize costs. First, densely connected convolutional network considered classification....

Point symbols on a scanned topographic map (STM) provide crucial geographic information. However, point symbol recognition entails high complexity and uncertainty owing to the stickiness of elements singularity structures. Therefore, extracting from STMs is challenging. Currently, performed primarily through pattern methods that have low accuracy efficiency. To address this problem, we investigated potential deep learning-based method for proposed convolutional neural network (DCNN)-based...

Building change detection (CD) using remote sensing images plays a vital role in urban development, and deep learning models attracted attention for their potential to accomplish CD tasks automatically. However, most methods are still facing challenges, such as the costly time-consuming process of constructing datasets severely imbalanced distribution positive negative samples preventing loss functions from functioning desired training process. Inspired by weak supervision have demonstrated...

Deep-learning based approaches have been proven effective for Digital Elevation Model (DEM) super-resolution (SR) tasks. Previous networks typically treat DEM elevation values as single-channel image input. However, images alone cannot fully capture spatial and terrain features. Shaded relief (SRIs), derived from DEMs, serve crucial visual cues that intuitively convey characteristics, addressing the limitations of providing synergistic benefits training DL models. The primary challenge in...

Cross-domain classification with small samples is a more challenging and realistic experimental setup. Until now, few studies have focused on the problem of small-sample cross-domain between completely different hyperspectral images (HSIs) since they possess land cover types statistical characteristics. To this end, paper proposes general-purpose representation learning method for HSI classification, aiming to enable model learn deep representations that can quickly adapt target domains...

The building extraction from remote sensed images ash been a challenging yet vital task for applicable purposes such as urban monitoring and cartography. Most of the existing learning based approaches focus on supervised methods, which models should be trained with corresponding labels. This research exploits self-supervised approach extraction, could train backbone within network without annotations. Specifically, is initially pixel-level module instead commonly used or instance-level...

Deep learning based methods have recently been successfully explored in hyperspectral image classification field. However, training a deep model still requires large number of labeled samples, which is usually impractical images. In this paper, simple but effective feature extraction method proposed for classification. Specifically, pretrained convolutional neural network on the ImageNet dataset used to extract spatial features image. Recently, it easy obtain Internet. Note that models are...

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI