Semantic segmentation of remote sensing images plays an important role in a wide range applications including land resource management, biosphere monitoring and urban planning. Although the accuracy semantic has been increased significantly by deep convolutional neural networks, several limitations exist standard models. First, for encoder-decoder architectures such as U-Net, utilization multi-scale features causes underuse information, where low-level high-level are concatenated directly...

In recent years, researchers have paid increasing attention on hyperspectral image (HSI) classification using deep learning methods. To improve the accuracy and reduce training samples, we propose a double-branch dual-attention mechanism network (DBDA) for HSI in this paper. Two branches are designed DBDA to capture plenty of spectral spatial features contained HSI. Furthermore, channel block applied these two respectively, which enables refine optimize extracted feature maps. A series...

The fully convolutional network (FCN) with an encoder-decoder architecture has been the standard paradigm for semantic segmentation. utilizes encoder to capture multilevel feature maps, which are incorporated into final prediction by a decoder. As context is crucial precise segmentation, tremendous effort made extract such information in intelligent fashion, including employing dilated/atrous convolutions or inserting attention modules. However, these endeavors all based on FCN ResNet other...

Semantic segmentation of remotely sensed imagery plays a critical role in many real-world applications, such as environmental change monitoring, precision agriculture, protection, and economic assessment. Following rapid developments sensor technologies, vast numbers fine-resolution satellite airborne remote sensing images are now available, for which semantic is potentially valuable method. However, because the rich complexity heterogeneity information provided with an ever-increasing...

The attention mechanism can refine the extracted feature maps and boost classification performance of deep network, which has become an essential technique in computer vision natural language processing. However, memory computational costs dot-product increase quadratically with spatiotemporal size input. Such growth hinders usage mechanisms considerably application scenarios large-scale inputs. In this letter, we propose a linear (LAM) to address issue, is approximately equivalent...

Semantic segmentation from very fine resolution (VFR) urban scene images plays a significant role in several application scenarios including autonomous driving, land cover classification, planning, etc. However, the tremendous details contained VFR image, especially considerable variations scale and appearance of objects, severely limit potential existing deep learning approaches. Addressing such issues represents promising research field remote sensing community, which paves way for...

The thriving development of earth observation technology makes more and high-resolution remote-sensing images easy to obtain. However, caused by fine-resolution, the huge spatial spectral complexity leads automation semantic segmentation becoming a challenging task. Addressing such an issue represents exciting research field, which paves way for scene-level landscape pattern analysis decision-making. To tackle this problem, we propose approach automatic land based on Feature Pyramid Network...

In this paper, a Multi-Scale Fully Convolutional Network (MSFCN) with multi-scale convolutional kernel is proposed to exploit discriminative representations from two-dimensional (2D) satellite images.

Semantic segmentation of remotely sensed images plays an important role in land resource management, yield estimation, and economic assessment. U-Net, a deep encoder–decoder architecture, has been used frequently for image with high accuracy. In this letter, we incorporate multiscale features generated by different layers U-Net design skip connected asymmetric-convolution-based (MACU-Net), using fine-resolution images. Our the following advantages: (1) connections combine realign...

In recent years, researchers have paid increasing attention on hyperspectral image (HSI) classification using deep learning methods. To improve the accuracy and reduce training samples, we propose a double-branch dual-attention mechanism network (DBDA) for HSI in this paper. Two branches are designed DBDA to capture plenty of spectral spatial features contained HSI. Furthermore, channel block applied these two respectively, which enables refine optimize extracted feature maps. A series...

In remote sensing images, the presence of thick cloud accompanying shadow is a high probability event, which can affect quality subsequent processing and limit scenarios application. Hence, removing as well recovering cloud-contaminated pixels indispensable to make good use images. this paper, novel removal method for images based on temporal smoothness sparsity-regularized tensor optimization (TSSTO) proposed. The basic idea TSSTO that are not only sparse but also smooth along horizontal...

Clouds in remote sensing optical images often obscure essential information. They may lead to occlusion or distortion of ground features, thereby affecting the subsequent analysis and extraction target Therefore, removal clouds is a critical task various applications. SAR-optical image fusion has achieved encouraging performance reconstruction cloud-covered Such methods, however, are extremely time-consuming computationally intensive, making them difficult apply practice. This letter...

In this paper, to remedy deficiency, we propose a Linear Attention Mechanism which is approximate dot-product attention with much less memory and computational costs. The efficient design makes the incorporation between mechanisms neural networks more flexible versatile. Experiments conducted on semantic segmentation demonstrated effectiveness of linear mechanism. Code available at https://github.com/lironui/Linear-Attention-Mechanism.

Assigning geospatial objects with specific categories at the pixel level is a fundamental task in remote sensing image analysis. Along rapid development of sensor technologies, remotely sensed images can be captured multiple spatial resolutions (MSR) information content manifested different scales. Extracting from these MSR represents huge opportunities for enhanced feature representation and characterisation. However, suffer two critical issues: (1) increased scale variation geo-objects (2)...

Presence of cloud-covered pixels is inevitable in optical remote-sensing images. Therefore, the reconstruction details important to improve usage these images for subsequent image analysis tasks. Aiming tackle issue high computational resource requirements that hinder application at scale, this paper proposes a Feature Enhancement Network(FENet) removing clouds satellite by fusing Synthetic Aperture Radar (SAR) and The proposed network consists designed Aggregation Residual Block...

Hyperspectral Image (HSI) classification based on deep learning has been an attractive area in recent years. However, as a kind of data-driven algorithm, method usually requires numerous computational resources and high-quality labelled dataset, while the cost high-performance computing data annotation is expensive. In this paper, to reduce dependence massive calculation samples, we propose lightweight network architecture (LiteDenseNet) DenseNet for Classification. Inspired by GoogLeNet...

In remote sensing images, the existence of thin cloud is an inevitable and ubiquitous phenomenon that crucially reduces quality imageries limits scenarios application. Therefore, removal indispensable procedure to enhance utilization images. Generally, even though contaminated by clouds, pixels still retain more or less surface information. Hence, different from thick removal, algorithms normally concentrate on inhibiting influence rather than substituting cloud-contaminated pixels....

Due to the high sensitivity and fast response, light‐screen array measurement principle is suitable for dynamic parameter of small targets including projectile. Since spatial structures determine accuracy, internal parameters such as angles between light‐screens are usually calibrated then directly used in field. However, effect measuring state ignored test This paper takes integrated sky vertical target research object, two rotation introduced external describe deviation calibration target,...

In multi-view reconstruction, mesh models are one of the most important data carriers. However, existing reconstruction methods have not adequately considered existence real-world edge features, resulting in insufficient utilization geometric feature information from images. This leads to issues such as surface distortion and a lack prominent features reconstructed models. To address these challenges, this paper proposes method line expression during process. algorithm integrates image model...