Remote sensing scene classification (RSSC) is a hotspot and play very important role in the field of remote image interpretation recent years. With development convolutional neural networks, significant breakthrough has been made scenes. Many objects form complex diverse scenes through spatial combination association, which makes it difficult to classify The problem insufficient differentiation feature representations extracted by Convolutional Neural Networks (CNNs) still exists, mainly due...

Image segmentation has made great progress in recent years, but the annotation required for image is usually expensive, especially remote sensing images. To solve this problem, we explore semi-supervised learning methods and appropriately utilize a large amount of unlabeled data to improve performance segmentation. This paper proposes method based on learning. We first design Consistency Regularization (CR) training training, then employ new learned model Average Update Pseudo-label (AUP),...

With the development of deep learning, remote sensing (RS) image segmentation has been applied with marked success. However, in process model training, large number labeled images required more expensive annotation. A key challenge is how to make full use extensive unlabeled available improve model. In this article, we propose a semisupervised semantic method defined as RanPaste, which combines performance. First, obtain pseudo label by randomly pasting part ground truth into predicted map....

With the development of deep learning, semantic segmentation remote sensing images has made great progress. However, algorithms based on learning usually require a huge number labeled for model training. For images, pixel-level annotation consumes expensive resources. To alleviate this problem, letter proposes semi-supervised method an iterative contrastive network. This combines few and more unlabeled to significantly improve performance. First, networks continuously learn potential...

With the development of deep learning in remote sensing (RS) image change detection (CD), dependence CD models on labeled data has become an important problem. To make better use comparatively resource-saving unlabeled data, method based semi-supervised (SSL) is worth further study. This article proposes a reliable contrastive (RCL) for RS CD. First, according to task characteristics CD, we design loss changed areas enhance model's feature extraction ability objects. Then, improve quality...

Although neural networks have achieved great success in various fields, applications on mobile devices are limited by the computational and storage costs required for large models. The model compression (neural network pruning) technology can significantly reduce parameters improve efficiency. In this article, we propose a differentiable channel pruning (DNCP) method compression. Unlike existing methods that require sampling evaluation of number substructures, our efficiently search optimal...

The past several years have witnessed the rapid development of task road extraction in high-resolution remote sensing images. However, due to complex background and distribution, is still a challenging research In convolutional neural networks (CNNs), U-shaped architecture network has shown its effectiveness. But global representation cannot be captured effectively by CNNs. While transformer, self-attention (SA) module can capture long-distance feature dependencies. A hybrid encoder-decoder...

Rapid progress has been made in the research of high-resolution remote sensing road extraction tasks past years, but due to diversity types and complexity context, extracting perfect network is still fraught with difficulties challenges. Many Convolutional Neural Networks (CNNs) based on encoder-decoder structures have demonstrated their effectiveness. Transformer's self-attention mechanism shows more powerful performance than CNNs modeling global feature dependencies. In this paper, we...

Building extraction from high spatial resolution remote sensing images is a hot spot in the field of applications and computer vision. This paper presents semantic segmentation model, which supervised method, named Pyramid Self-Attention Network (PISANet). Its structure simple, because it contains only two parts: one backbone network, used to learn local features (short distance context information around pixel) buildings image; other part pyramid self-attention module, obtain global (long...

Although widely exploited in recent decades, road extraction is still a very significant and challenging research the field of remote sensing image processing due to complex background distribution. Among existing CNN-based methods, U-shape architectures composed encoders decoders have shown their effectiveness. In this letter, we propose an improved encoder–decoder method, named DBRANet, for extracting roads from images. encoding phase, present dual-branch network module (DBNM) construct...

Semantic segmentation is a basic task in computer vision, but only limited attention has been devoted to the ultra-high-resolution (UHR) image segmentation. Since UHR images occupy too much memory, they cannot be directly put into GPU for training. Previous methods are cropping small patches or downsampling whole images. Cropping and cause loss of contexts details, which essential accuracy. To solve this problem, we improve simplify local global feature fusion method previous works. Local...

Remote sensing (RS) images present unique challenges for computer vision due to lower resolution, smaller objects, and fewer features. Mainstream backbone networks show promising results traditional visual tasks. However, they use convolution reduce feature map dimensionality, which can result in information loss small objects RS decreased performance. To address this problem, we propose a new universal downsampling module named Robust Feature Downsampling (RFD). RFD fuses multiple maps...

Remote sensing (RS) visual tasks have gained significant academic and practical importance. However, they encounter numerous challenges that hinder effective feature extraction, including the detection recognition of multiple objects exhibiting substantial variations in scale within a single image. While prior dual-branch or multi-branch architectural strategies been managing these object variances, concurrently resulted considerable increases computational demands parameter counts....

Without constructing adjacency graph for neighborhood, we propose a method to learn similarity among sample points of manifold in Laplacian embedding (LE) based on adding constraints linear reconstruction and least absolute shrinkage selection operator type minimization. Two algorithms corresponding analyses are presented mix-signed nonnegative data respectively. The learning is further extended kernel spaces. experiments both synthetic real world benchmark sets demonstrate that the proposed...