Abstract:The low spatial resolution of hyperspectral images leads to the coexistence of multiple ground objects in a single pixel (called mixed pixels). A large number of mixed pixels in a hyperspectral image hinders the subsequent analysis and application of the image. In order to solve this problem, a novel sparse unmixing method, which considers highly similar patches in nonlocal regions of a hyperspectral image, is proposed in this article. This method exploits spectral correlation by using collaborative sparsity r… Show more
“…Sun et al [62] propose a novel patch-based low rank component induced spatial-spectral kernel method for hyperspectral image (HSI) classification. In [63], a novel sparse unmixing method is proposed for hyperspectral image classification, which utilizes spectral correlation by using collaborative sparsity regularization and weighted nonlocal low-rank tensor regularization. Cheng et al [64] propose an auto-encoder pair model for visual tracking which is composed of source domain network and target domain network to help a more accurate localization.…”
With the amount of remote sensing data increasing at an extremely fast pace, machine learning based technique has been shown to perform superiorly in many applications. However, most of existing methods in real-time application are based on single modal image data. Although a few approaches use the different source images to represent the object via fusion scheme, it may not be appropriate for multi-modality information processing. In addition, these methods hardly benefit from the end-to-end network training due to the limitations of implementation difficulty and computational cost. In this paper, we propose a multi-task multi-source information fusion method in the deep learning and correlation filter frameworks, which is applied to the fields of tracking and remote sensing data processing. The contribution of individual layers from different source data inside the deep network model is considered as a task. The proposed method can employ interdependencies among different sources data and tasks to learn deep network parameters and filters jointly to improve the performance. Second, we present an effective object appearance selection scheme to adaptively capture the object appearance changes via effective deep learning network, then integrating information from different modalities to achieve information fusion. Different source information can provide robust performance from different aspects with complementary properties. Third, we further extend the proposed approach to the field of remote sensing for semantic labeling. The layers' sensitivity is utilized to verify the robustness of different classes. Extensively experiments on five benchmarks show that the proposed approach performs favorably against the state-of-the-arts.
“…Sun et al [62] propose a novel patch-based low rank component induced spatial-spectral kernel method for hyperspectral image (HSI) classification. In [63], a novel sparse unmixing method is proposed for hyperspectral image classification, which utilizes spectral correlation by using collaborative sparsity regularization and weighted nonlocal low-rank tensor regularization. Cheng et al [64] propose an auto-encoder pair model for visual tracking which is composed of source domain network and target domain network to help a more accurate localization.…”
With the amount of remote sensing data increasing at an extremely fast pace, machine learning based technique has been shown to perform superiorly in many applications. However, most of existing methods in real-time application are based on single modal image data. Although a few approaches use the different source images to represent the object via fusion scheme, it may not be appropriate for multi-modality information processing. In addition, these methods hardly benefit from the end-to-end network training due to the limitations of implementation difficulty and computational cost. In this paper, we propose a multi-task multi-source information fusion method in the deep learning and correlation filter frameworks, which is applied to the fields of tracking and remote sensing data processing. The contribution of individual layers from different source data inside the deep network model is considered as a task. The proposed method can employ interdependencies among different sources data and tasks to learn deep network parameters and filters jointly to improve the performance. Second, we present an effective object appearance selection scheme to adaptively capture the object appearance changes via effective deep learning network, then integrating information from different modalities to achieve information fusion. Different source information can provide robust performance from different aspects with complementary properties. Third, we further extend the proposed approach to the field of remote sensing for semantic labeling. The layers' sensitivity is utilized to verify the robustness of different classes. Extensively experiments on five benchmarks show that the proposed approach performs favorably against the state-of-the-arts.
“…Using the idea of increasing additional constraints, Feng et al [18] improve the plain MVNTF method by integrating sparseness, volume, and nonlinearity constraints into the cost function. Besides, lowrank constraints for abundance and endmember tensors have also been adopted in NTF-based unmixing methods [19], [20].…”
Spectral unmixing is an important technique for quantitatively analyzing hyperspectral remote sensing images. Recently, constrained nonnegative matrix factorization (NMF) has been demonstrated to be a powerful tool for spectral unmixing. However, acquiring the problem-dependent prior knowledge and incorporating it into NMF as effective constraints is a challenging task. In this paper, a multiple clustering guided NMF (MCG-NMF) unmixing approach is proposed under a selfsupervised framework which has been used to effectively learn high-level semantic information from the data with a surrogate task in many applications. Specifically, in order to provide selfsupervised information to guide the NMF-based unmixing model, multiple clustering is integrated into the optimization process of NMF. Moreover, by introducing interaction between each clustering and the unmixing procedure, more accurate proximate endmember signatures and proximate abundance distributions are expected to be acquired and used to impose self-supervised constraints on endmembers and abundances respectively. Consequently, effective prior information about endmember signatures and abundance distributions is captured and simultaneously integrated into NMF as valuable constraints to promote unmixing performance. Experiments are conducted on both synthetic data and real HSIs, and the superior performance of our method is shown by comparing it with several state-of-the-art algorithms.
“…At present, the structure-based image inpainting [1][2][3][4], texture-based image inpainting [5][6][7][8][9][10], and deep learning-based image inpainting [11][12][13][14][15][16] are the three main directions in the research field of image inpainting. The research in the paper is mainly aimed at image learning algorithms based on deep learning.…”
Various problems existed in the image inpainting algorithms, which can't meet people's requirements visually. Aiming at the defects of the existing image inpainting algorithms, such as low accuracy, poor visual consistency, and unstable training, an improved image inpainting algorithm used on a multi-scale generative adversarial network (GAN) and neighbourhood model have been proposed in the paper. The proposed algorithm mainly improves the network structure of the discriminator, and introduces a multi-scale discriminator based on the global discriminator and the local discriminator. The multi-scale discriminators were trained on images of different resolutions. Discriminators of different scales have different receptive fields, which can guide the generator to generate more global image views and finer details. Aiming at the problem of gradient disappearance or gradient explosion that often occurs in GAN training, the method of WGAN (Wasserstein GAN) has been used to simulate the sample data distribution using EM distance. The proposed model has been trained and tested on the CelebA, ImageNet, and Place2. The experimental results show that compared with the previous algorithm model, the proposed algorithm improves the accuracy of image inpainting and can generate more realistic repairing images, and it is suitable for many types of images.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.