Semantic segmentation is an extremely challenging task in high-resolution remote sensing (HRRS) images as objects have complex spatial layouts and enormous variations appearance. Convolutional neural networks (CNNs) excellent ability to extract local features been widely applied the feature extractor for various vision tasks. However, due inherent inductive bias of convolution operation, CNNs inevitably limitations modeling long-range dependencies. Transformer can capture global...

To mine the spectral-spatial information of target pixel in hyperspectral image classification (HSIC), convolutional neural network (CNN)-based models widely adopt patch-based input pattern, where a patch represents its central and neighbor pixels play auxiliary roles process. However, compared to pixel, often have different contributions for classification. Although many existing CNNs could adaptively emphasize spatial information, most them ignore latent relationship between center pixels....

For the abundant spectral and spatial information recorded in hyperspectral images (HSIs), fully exploring spectral-spatial relationships has attracted widespread attention image classification (HSIC) community. However, there are still some intractable obstructs. one thing, patch based processing pattern, neighbor pixels often inconsistent with central pixel land-cover class. another linear nonlinear correlations between different bands vital yet tough for representing excavating. To...

Image reconstruction-based methods with autoencoder have been widely used for unsupervised anomaly detection. By training the reconstruction on normal samples, is supposed to produce higher error anomalous which as an indicator detecting anomalies. However, since adopts bottleneck layer reconstruct data, it hard control its generalization capability. When capability high, features can be confused features, resulting in accurate of regions well. In this article, we propose a dual-constraint...

Semantic segmentation for remote sensing is a crucial but challenging task. Many supervised semantic methods rely heavily on large-scale pixel-wise annotated data set, it time-consuming and laborious to provide manual annotation. However, due the common domain shift of images, direct transfer might not perform well. Therefore, many unsupervised adaptation have been proposed solve distribution discrepancy in remote-sensing sets, these cannot completely utilize features extracted training...

Remote sensing change detection refers to the process of identifying and extracting changes in objects within same geographical region over multiple periods. With increasing spatial resolution remote images, minor has become a challenging task. We introduce multilevel feature aggregation enhancement network tackle this issue. Specifically, we propose module aggregate distinct features extracted from each image, which strengthens representation capability. Subsequently, difference parallel...

With the advancement of satellite technology, application space change detection (CD) in remote sensing images is continuously expanding. However, development technology still ongoing, and limited resolution complex ground object information remain significant challenges field CD. Recent CD networks generally utilize multi-feature fusion to make full use detailed at different scales. most have capabilities handling large-scale feature maps, leading an impact on effectiveness detecting...

Surface defect detection is a significant step in industrial production, which also essential for ensuring the quality of products. At moment, although methods based on computer vision have made great progress, they are still tough to automate due challenges large changes size, low contrast, strong background interference, intraclass difference as well small interclass difference. To address above issues, we design deep learning model that mainly consists feature augmentation module (FAM),...

Deep convolutional neural networks have made significant progress in the field of intelligent analysis remote-sensing images. However, semantic segmentation task high-resolution (HRRS) images always faces problem large-scale variation and complex background samples, which causes difficulties distinguishing confusable ground objects. In this letter, we propose a novel multipath feature refinement network (MFRNet) to alleviate above problems. We design module (FRM) fuse features at various...

The goal of unsupervised surface anomaly detection is to detect areas the image that are different from normal pattern, which can be considered as a semantic segmentation problem oriented anomalous patterns. However, this challenging due lack actual available samples. In paper, we transform into self-supervised by proposed simulation strategy. Using only samples for training, real anomalies appearing in inference phase detected. Thus, propose network with joint representation and contrast...

In this paper, we present an efficient network to tackle three critical problems in high spatial resolution (HSR) remote sensing image segmentation: (i) feature misalignment, (ii) insufficient contextual information extraction and (iii) various class imbalance issues. detail, propose a novel Feature Alignment Block (FAB) suppress misalignment issues with the guide of anchor map. Further, extract sufficient information, design Contextual Augmentation (CAB) augment features different semantic...

Change detection (CD) plays a critical role in extracting ground changes from bi-temporal remote sensing (RS) images and is instrumental understanding surface dynamics. In recent years, deep learning has made significant breakthroughs CD. However, typical CD methods that employ the Siamese network for temporal feature extraction lack alignment ability heterogeneous RS images, resulting inadequate discriminative capability. Moreover, learning-based are still susceptible to problem of minor...

Camouflage is a natural or artificially process to prevent an object from being detected while camouflage breaking countering for the identification of concealed object. We report that perfectly camouflaged in two-dimensional scene can be retrieved and with stereo-vision assisted three-dimensional (3D) imaging perceived stereopsis. The analysis based on binocular energy model applied general 3D settings. show random noise background vision’s stereoacuity resolve hidden structures....

Semantic segmentation of high spatial resolution (HSR) remote sensing images (RSIs) plays an important role in many applications. However, HSR RSIs have significantly larger sizes than typical natural images, which results fewer valuable samples when training models. In addition, fusing multiscale features is the key step obtaining with strong semantic and information. current feature fusion methods are too straightforward to address misalignment issues. To handle these two problems, we...