Quality assessment of omnidirectional images has become increasingly urgent due to the rapid growth of virtual reality applications. Different from traditional 2D images and videos, omnidirectional contents can provid...
详细信息
Rain removal is important for many computer vision applications, such as surveillance, autonomous car, etc. Traditionally, rain removal is regarded as a signal removal problem which usually causes over-smoothing by re...
详细信息
Supraglacial lake plays an important role in ice sheet dynamics, mass balance and sea level rise. Therefore, it is of great importance to extract supraglacial lake and obtain its spatial-temporal distribution or chang...
详细信息
ISBN:
(纸本)9781665468893
Supraglacial lake plays an important role in ice sheet dynamics, mass balance and sea level rise. Therefore, it is of great importance to extract supraglacial lake and obtain its spatial-temporal distribution or change. This study provides an automatic extraction model for supraglacial lake, using Synthetic Aperture Radar (SAR) imagery based on deep learning. First, select 19580 Sentinel-1 SAR imagery patches in eight typical areas for manual labeling. Second, the GPU-based U-Net model is used to implement the training of the supraglacial lake, and the results are evaluated in different sites. Finally, the training model is used to perform the supraglacial lake extraction. In addition, this article also introduces ArcticDEM to remove shadow confusion in the margin of the ice sheet. The global-local threshold segmentation method is used to extract the supraglacial lake on the Sentinel-2 MSI imagery, which is a comparative analysis and information supplement for the extracted results in this paper. The results show that: (1) The U-Net network selected in this paper is suitable for processing small sample size SAR data and multi-modal feature extraction. The GPU parallel processing method can achieve rapid extraction of massive data and reduce time cost. (2) The Dice coefficient of the training model reaches 0.98, which can be used for effective extraction of supraglacial lake. (3) Compared with the results of optical image extraction, the algorithm proposed in this paper can identify lakes in areas covered by snow or thin ice, which truly reflects the supraglacial lake temporal and spatial distribution characteristics.
Spaceborne Interferometric Synthetic Aperture Radar (InSAR) has the capability of high precise topographic mapping for large area. However, on the one hand, digital elevation models (DEM) inversion needs at least one ...
详细信息
Generative adversarial nets (GANs) have been successfully applied in many fields like image generation, inpainting, super-resolution, and drug discovery, etc. By now, the inner process of GANs is far from being unders...
详细信息
Generative adversarial nets (GANs) have been successfully applied in many fields like image generation, inpainting, super-resolution, and drug discovery, etc. By now, the inner process of GANs is far from being understood. To get a deeper insight into the intrinsic mechanism of GANs, in this paper, a method for interpreting the latent space of GANs by analyzing the correlation between latent variables and the corresponding semantic contents in generated images is proposed. Unlike previous methods that focus on dissecting models via feature visualization, the emphasis of this work is put on the variables in latent space, i.e. how the latent variables affect the quantitative analysis of generated results. Given a pre-trained GAN model with weights fixed, the latent variables are intervened to analyze their effect on the semantic content in generated images. A set of controlling latent variables can be derived for specific content generation, and the controllable semantic content manipulation is achieved. The proposed method is testified on the datasets Fashion-MNIST and UT Zappos50K, experiment results show its effectiveness.
Single image super-resolution (SISR) algorithms reconstruct high-resolution (HR) images with their low-resolution (LR) counterparts. It is desirable to develop image quality assessment (IQA) methods that can not only ...
详细信息
Convolutional neural networks (CNNs) have made tremendous success in optical images classification recently. However, in synthetic aperture radar (SAR) target classification, it is difficult to annotate a large amount...
详细信息
ISBN:
(纸本)9781665468893
Convolutional neural networks (CNNs) have made tremendous success in optical images classification recently. However, in synthetic aperture radar (SAR) target classification, it is difficult to annotate a large amount of real SAR images to train CNNs. Sufficient annotated images can be easily obtained through simulation, but the disparity between the simulated images and the real images makes them difficult to directly apply to the real images classification. In this paper, we propose a model that integrates multi-kernel maximum mean discrepancy (MK-MMD) and domain-adversarial training to alleviate this problem. Simulated SAR images with annotation and unlabeled real SAR images are used to train our model. First, we use domain-adversarial training to prompt the model to extract domain-invariant features. Then, the MK-MMD between the hidden representations of simulated images and real images is reduced to narrow domain discrepancy. Experimental results on the real SAR dataset demonstrate that our method effectively solves the domain shift problem and improves the classification accuracy.
SyntheticAperture Radar (SAR) is an active microwave imaging system, which can provide all-time and allweather imaging with a wide observation range. In sub-meter high-resolution SAR images, man-made metallic targets,...
详细信息
ISBN:
(纸本)9781728173344
SyntheticAperture Radar (SAR) is an active microwave imaging system, which can provide all-time and allweather imaging with a wide observation range. In sub-meter high-resolution SAR images, man-made metallic targets, such as pylons, can be easily detected and their details can be obtained. Therefore, pylon detection using high-resolution SAR images has its unique advantages. To solve the problem that traditional pylon detection methods in SAR images require the manual design of feature extractors and perform poorly in terms of realtime requirements, an effective and fast framework for pylon detection in SAR images is proposed in this paper. It combines Faster R-CNN and sliding window algorithm to detect pylons in large-scene SAR images. Besides, it effectively improves detection accuracy by using data augmentation. The proposed method can detect up to 97.6% on the test set and 80% in largescene GF-3 SAR images. The average time cost of detection is 0. 06s for a slice image and 90s for a large-scale SAR image. Experiments demonstrate good performance in detection rate and instantaneity.
Deep learning based patch-wise Synthetic Aperture Radar (SAR) image classification usually requires a large number of labeled data for training. Aiming at understanding SAR images with very limited annotation and taki...
详细信息
ISBN:
(数字)9781728163741
ISBN:
(纸本)9781728163758
Deep learning based patch-wise Synthetic Aperture Radar (SAR) image classification usually requires a large number of labeled data for training. Aiming at understanding SAR images with very limited annotation and taking full advantage of complex-valued SAR data, this paper proposes a general and practical framework for quad-, dual-, and single-polarized SAR data. In this framework, two important elements are taken into consideration: image representation and physical scattering properties. Firstly, a convolutional neural network is applied for SAR image representation. Based on time-frequency analysis and polarimetric decomposition, the scattering labels are extracted from complex SAR data with unsupervised deep learning. Then, a bag of scattering topics for a patch is obtained via topic modeling. By assuming that the generated scattering topics can be regarded as the abstract attributes of SAR images, we propose a soft constraint between scattering topics and image representations to refine the network. Finally, a classifier for land cover and land use semantic labels can be learned with only a few annotated samples. The framework is hybrid for the combination of deep neural network and explainable approaches. Experiments are conducted on Gaofen-3 complex SAR data and the results demonstrate the effectiveness of our proposed framework.
Synthetic aperture radar (SAR) tomography (TomoSAR) is a novel technique that enables three-dimensional (3-D) imaging and plays an important role in urban remote sensing by utilizing multiple observations of the same ...
详细信息
ISBN:
(纸本)9781665468893
Synthetic aperture radar (SAR) tomography (TomoSAR) is a novel technique that enables three-dimensional (3-D) imaging and plays an important role in urban remote sensing by utilizing multiple observations of the same target scene from various baselines. Canonical TomoSAR observations are from a single aspect, which has been well studies already. However, modern SAR sensors such as Unmanned Aerial Vehicle (UAV) allow us to achieve multi-aspect TomoSAR data of the same target scene. This paper proposes a novel framework named “Multi-aspect TomoSAR,” which takes advantage of rich TomoSAR data from multiple observation aspects. We derive the multi-aspect TomoSAR signal model using distributed compressed sensing (DCS) and adopt a simultaneous sparse approximation algorithm named SOMP to solve the joint sparsity model. Numerical results on synthetic simulated data show that the multi-aspect estimation can provide more accurate estimation, yield a promising perspective. Experimental results on real airborne data will be reported in the journal version of this work later.
暂无评论