Search

Recently, graph learning-based hyperspectral band selection algorithms illustrate impressive performance for hyperspectral image (HSI) processing, whose goal is to select an optimal band combination containing less redundancy through the learned graph matrix. However, most of the previous methods work in single spectral domain, which neglects the rich image spatial information. Moreover, they typically model the graph matrix from a global perspective while ignoring the differences in spatial distribution that exist in diverse pixel and band regions. Based on the above considerations, to take full account of structure information in spatial and spectral domains, a structural graph learning method for hyperspectral band selection (SGLM) is designed. Specifically, SGLM constructs two matrices using the correlation between pixels and bands under the local perspective, which can capture the image structure information reasonably. Since the proposed method is modelled in both spatial and spectral dimensions, it is conducive to subsequent band subset generation task. Meanwhile, in order to guarantee local spatial consistency among bands, a Laplacian regularization term associated with the error matrix is introduced to the self-representation model. Additionally, considering that the importance of a band is consistent with its capability to reconstruct the entire band set, an adjacent bands reconstruction strategy is adopted to obtain the ultimate band combination, which can assess the significance of each band effectively. The comprehensive experimental analysis on four datasets demonstrates that the proposed SGLM model has outstanding superiority in comparison with several band selection algorithms. [ABSTRACT FROM AUTHOR]

The knowledge graph (KG) is an efficient form of knowledge organization and expression, providing prior knowledge support for various downstream tasks, and has received extensive attention in natural language processing. However, existing large-scale KGs have many hidden facts that need to be discovered. How to effectively use the structure information of KG is an important research direction of knowledge reasoning. Structure-Information-based reasoning over the KG is a technique used to find the missing facts by the structure information of KG. This survey summarizes the methods and applications of Structure-Information-based reasoning and hopes to be helpful to the research in this field. First, we introduced the definition of knowledge reasoning and the conceptual description of related tasks. Then, we reviewed the methods of Structure-Information-based reasoning. Specifically, we categorized them into four representative classes: PRA-based reasoning, Path-Embedding-based reasoning, RL-based reasoning, and GNN-based reasoning. We compared the motivations and details between practices in the same category. After that, we described the application of Structure-Information-based knowledge reasoning in the KG Completion, Question Answering System, Recommendation System, and other fields. Finally, we discussed the future research directions of Structure-Information-based reasoning. [ABSTRACT FROM AUTHOR]

Objective quality assessment of three-dimensional (3D) point clouds is essential for the development of immersive multimedia systems in real-world applications. Despite the success of perceptual quality evaluation for 2D images and videos, blind/no-reference metrics are still scarce for 3D point clouds with large-scale irregularly distributed 3D points. Therefore, in this article, we propose an objective point cloud quality index with Structure Guided Resampling (SGR) to automatically evaluate the perceptually visual quality of dense 3D point clouds. The proposed SGR is a general-purpose blind quality assessment method without the assistance of any reference information. Specifically, considering that the human visual system is highly sensitive to structure information, we first exploit the unique normal vectors of point clouds to execute regional pre-processing that consists of keypoint resampling and local region construction. Then, we extract three groups of quality-related features, including (1) geometry density features, (2) color naturalness features, and (3) angular consistency features. Both the cognitive peculiarities of the human brain and naturalness regularity are involved in the designed quality-aware features that can capture the most vital aspects of distorted 3D point clouds. Extensive experiments on several publicly available subjective point cloud quality databases validate that our proposed SGR can compete with state-of-the-art full-reference, reduced-reference, and no-reference quality assessment algorithms. [ABSTRACT FROM AUTHOR]

Summary: Hierarchical classification learning aims to exploit the hierarchical relationship between data categories. The high dimensionality and dynamic of the data feature space are the main challenges of this research. Hierarchical feature selection uses a hierarchical structure to divide large‐scale tasks into multiple small tasks, which can more effectively improve the training speed and prediction accuracy of classification models. To present, existing online feature selection methods ignore the hierarchical structure of data. In addition, the dependency relationships in the hierarchical structure can serve as auxiliary knowledge to aid feature selection. Based on this, this paper proposes an online streaming hierarchical feature selection method based on kernelized fuzzy rough sets (OFS‐HNFRS). First, we use the prior knowledge of the hierarchical structure to divide the sample set into multiple subsets. Second, the dependency relationship of hierarchical structure is extended to kernelized fuzzy rough sets, and hierarchical category dependency based on kernelized fuzzy rough sets is defined. Finally, a new online feature selection framework is proposed, which is used to evaluate the relevance, significance, and redundancy of features. We verify the effectiveness of the proposed algorithm on six hierarchical datasets and eight flat datasets. [ABSTRACT FROM AUTHOR]

Closed-circuit television (CCTV) videos provide important data for intelligent control systems; however, sensitive information such as human faces and vehicle license plates must be de-identified by a de-identification framework before such videos are used. Previous research has mainly focused on enhancing the performance of de-identification methods, and only a few are devoted to detecting sensitive information. To address this issue, this study proposes an artificial intelligence-based sensitive information de-identification framework. The framework detects objects and utilizes their structural information to estimate the relative position of sensitive data, followed by the de-identification process. To enhance the recall of the detection module, this study analyzes difficult-to-detect faces and license plates in CCTV videos and proposes relative position estimation methods. These methods minimize the possibility of sensitive information leakage while reducing the number of locations of sensitive information that need to be annotated. When evaluated using actual urban CCTV videos, the recall of the framework was 84.63%, which is 33.01% higher than that of the conventional algorithm.

Closed-circuit television (CCTV) videos provide important data for intelligent control systems; however, sensitive information such as human faces and vehicle license plates must be de-identified by a de-identification framework before such videos are used. Previous research has mainly focused on enhancing the performance of de-identification methods, and only a few are devoted to detecting sensitive information. To address this issue, this study proposes an artificial intelligence-based sensitive information de-identification framework. The framework detects objects and utilizes their structural information to estimate the relative position of sensitive data, followed by the de-identification process. To enhance the recall of the detection module, this study analyzes difficult-to-detect faces and license plates in CCTV videos and proposes relative position estimation methods. These methods minimize the possibility of sensitive information leakage while reducing the number of locations of sensitive information that need to be annotated. When evaluated using actual urban CCTV videos, the recall of the framework was 84.63%, which is 33.01% higher than that of the conventional algorithm. [ABSTRACT FROM AUTHOR]

Abstract Ancient murals are precious cultural heritages. They suffer from various damages due to man-made destruction and long-time exposure to the environment. It is urgent to protect and restore the damaged ancient murals. Virtual restoration of ancient murals aims to fill damaged mural regions by using modern computer techniques. Most existing restoration approaches fail to fill the loss mural regions with rich details and complex structures. In this paper, we propose a virtual restoration network of ancient murals based on global–local feature extraction and structural information guidance (GLSI). The proposed network consists of two major sub-networks: the structural information generator (SIG) and the image content generator (ICG). In the first sub-network, SIG can predict the structural information and the coarse contents for the missing mural regions. In the second sub-network, ICG utilizes the predicted structural information and the coarse contents to generate the refined image contents for the missing mural regions. Moreover, we design an innovative BranchBlock module that can effectively extract and integrate the local and global features. We introduce a Fast Fourier Convolution (FFC) to improve the color restoration for the missing mural regions. We conduct experiments over simulated and real damaged murals. Experimental results show that our proposed method outperforms other three comparative state-of-the-art approaches in terms of structural continuity, color harmony and visual rationality of the restored mural images. In addition, the mural restoration results of our method can achieve comparatively high quantitative evaluation metrics.

Ancient murals are precious cultural heritages. They suffer from various damages due to man-made destruction and long-time exposure to the environment. It is urgent to protect and restore the damaged ancient murals. Virtual restoration of ancient murals aims to fill damaged mural regions by using modern computer techniques. Most existing restoration approaches fail to fill the loss mural regions with rich details and complex structures. In this paper, we propose a virtual restoration network of ancient murals based on global–local feature extraction and structural information guidance (GLSI). The proposed network consists of two major sub-networks: the structural information generator (SIG) and the image content generator (ICG). In the first sub-network, SIG can predict the structural information and the coarse contents for the missing mural regions. In the second sub-network, ICG utilizes the predicted structural information and the coarse contents to generate the refined image contents for the missing mural regions. Moreover, we design an innovative BranchBlock module that can effectively extract and integrate the local and global features. We introduce a Fast Fourier Convolution (FFC) to improve the color restoration for the missing mural regions. We conduct experiments over simulated and real damaged murals. Experimental results show that our proposed method outperforms other three comparative state-of-the-art approaches in terms of structural continuity, color harmony and visual rationality of the restored mural images. In addition, the mural restoration results of our method can achieve comparatively high quantitative evaluation metrics. [ABSTRACT FROM AUTHOR]

Camera-based indoor localization is a fundamental aspect of indoor navigation, virtual reality, and location-based services. Deep learning methods have exhibited remarkable performance with low storage requirements and high efficiency. However, existing methods mainly derive features implicitly for pose regression without considering explicit structure information from images. This paper proposes that incorporating such information can improve the localization performance of learning-based approaches. We extract structure information from RGB images in the form of depth maps and edge maps and design two modules for depth-weighted and edge-weighted feature fusion. These modules are integrated into the pose regression network to enhance pose prediction. Furthermore, we employ a self-attention module for high-level feature extraction to augment the network capacity. Extensive experiments are conducted on the publicly available 7Scenes and 12Scenes datasets, and the results demonstrate that the proposed method achieves high localization performance, with an average positional error of 0. 19 m and 0. 16 m , respectively. The code for this work is available at https://github.com/lqing900205/structureLoc. [ABSTRACT FROM AUTHOR]

Abstract Background As a highly aggressive disease, cancer has been becoming the leading death cause around the world. Accurate prediction of the survival expectancy for cancer patients is significant, which can help clinicians make appropriate therapeutic schemes. With the high-throughput sequencing technology becoming more and more cost-effective, integrating multi-type genome-wide data has been a promising method in cancer survival prediction. Based on these genomic data, some data-integration methods for cancer survival prediction have been proposed. However, existing methods fail to simultaneously utilize feature information and structure information of multi-type genome-wide data. Results We propose a Multi-type Data Joint Learning (MDJL) approach based on multi-type genome-wide data, which comprehensively exploits feature information and structure information. Specifically, MDJL exploits correlation representations between any two data types by cross-correlation calculation for learning discriminant features. Moreover, based on the learned multiple correlation representations, MDJL constructs sample similarity matrices for capturing global and local structures across different data types. With the learned discriminant representation matrix and fused similarity matrix, MDJL constructs graph convolutional network with Cox loss for survival prediction. Conclusions Experimental results demonstrate that our approach substantially outperforms established integrative methods and is effective for cancer survival prediction.

Graph-structured data, operating as an abstraction of data containing nodes and interactions between nodes, is pervasive in the real world. There are numerous ways dedicated to extract graph structure information explicitly or implicitly, but whether it has been adequately exploited remains an unanswered question. This work goes deeper by heuristically incorporating a geometric descriptor, the discrete Ricci curvature (DRC), in order to uncover more graph structure information. We present a curvature-based topology-aware graph transformer, termed Curvphormer. This work expands the expressiveness by using a more illuminating geometric descriptor to quantify the connections within graphs in modern models and to extract the desired structure information, such as the inherent community structure in graphs with homogeneous information. We conduct extensive experiments on a variety of scaled datasets, including PCQM4M-LSC, ZINC, and MolHIV, and obtain a remarkable performance gain on various graph-level tasks and fine-tuned tasks. [ABSTRACT FROM AUTHOR]

Despite the remarkable progress made in recent years, until today, the automatic detection of changes in synthetic aperture radar (SAR) images remains a difficult task due to speckle noise. This inherent multiplicative noise tends to increase false alarms and misdetections. As a solution, we developed an unsupervised method that detects SAR changes by analyzing structural differences. By this method, the spatial structure cues of a pixel are represented by a set of similarity weight vectors calculated from the non-local scale of the pixel. The difference image (DI) is then derived by measuring the structural consistency of the corresponding pixels. A new statistical distance that is insensitive to speckle noise was used to measure the similarity weights between patches in order to obtain an accurate structure. It was derived by applying the Nakagami–Rayleigh distribution to a statistical test and customizing the approximation based on change detection. The CFAR threshold estimator in conjunction with the Rayleigh hypothesis was then employed to attenuate the effect of the unimodal histogram of the DI. The results indicated that the proposed method reduces the false alarm rate and improves the kappa and F1-scores, while providing satisfactory visual results. [ABSTRACT FROM AUTHOR]

Aiming at the problems of patch-based synthesis algorithm of structured texture images,such as discontinuity of structure,distortion of boundary,seam misalignment,and low synthesis speed,a new fast non-overlapping synthesis algorithm of texture images is proposed based on the consistency criterion of double-seam lines,thereby effectively improving the synthesis quality and speed of structured texture images.Firstly,the seamline consistency criterion considering hue,saturation,brightness and edge characteristics simultaneously is established in HSI color space that is more consistent with human visual characteristic.Then,a sub-block search strategy and a new non-overlapping splicing algorithm based on the consistency criterion of double-seam line are proposed and implemented.The experiment results show that the proposed algorithm can significantly improve the synthesis quality and speed of structured texture images in comparison with the traditional algorithms.

Background: As a highly aggressive disease, cancer has been becoming the leading death cause around the world. Accurate prediction of the survival expectancy for cancer patients is significant, which can help clinicians make appropriate therapeutic schemes. With the high-throughput sequencing technology becoming more and more cost-effective, integrating multi-type genome-wide data has been a promising method in cancer survival prediction. Based on these genomic data, some data-integration methods for cancer survival prediction have been proposed. However, existing methods fail to simultaneously utilize feature information and structure information of multi-type genome-wide data. Results: We propose a Multi-type Data Joint Learning (MDJL) approach based on multi-type genome-wide data, which comprehensively exploits feature information and structure information. Specifically, MDJL exploits correlation representations between any two data types by cross-correlation calculation for learning discriminant features. Moreover, based on the learned multiple correlation representations, MDJL constructs sample similarity matrices for capturing global and local structures across different data types. With the learned discriminant representation matrix and fused similarity matrix, MDJL constructs graph convolutional network with Cox loss for survival prediction. Conclusions: Experimental results demonstrate that our approach substantially outperforms established integrative methods and is effective for cancer survival prediction. [ABSTRACT FROM AUTHOR]

Copyright of Frontiers of Information Technology & Electronic Engineering is the property of Springer Nature and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Deep self-expressiveness-based subspace clustering methods have demonstrated effectiveness. However, existing works only consider the attribute information to conduct the self-expressiveness, limiting the clustering performance. In this paper, we propose a novel adaptive attribute and structure subspace clustering network (AASSC-Net) to simultaneously consider the attribute and structure information in an adaptive graph fusion manner. Specifically, we first exploit an auto-encoder to represent input data samples with latent features for the construction of an attribute matrix. We also construct a mixed signed and symmetric structure matrix to capture the local geometric structure underlying data samples. Then, we perform self-expressiveness on the constructed attribute and structure matrices to learn their affinity graphs separately. Finally, we design a novel attention-based fusion module to adaptively leverage these two affinity graphs to construct a more discriminative affinity graph. Extensive experimental results on commonly used benchmark datasets demonstrate that our AASSC-Net significantly outperforms state-of-the-art methods. In addition, we conduct comprehensive ablation studies to discuss the effectiveness of the designed modules. The code is publicly available at https://github.com/ZhihaoPENG-CityU/AASSC-Net. [ABSTRACT FROM AUTHOR]

Visible-infrared person re-identification (VI-ReID) is a cross-modality retrieval problem, which aims at matching the same pedestrian between the visible and infrared cameras. Due to the existence of pose variation, occlusion, and huge visual differences between the two modalities, previous studies mainly focus on learning image-level shared features. Since they usually learn a global representation or extract uniformly divided part features, these methods are sensitive to misalignments. In this paper, we propose a structure-aware positional transformer (SPOT) network to learn semantic-aware sharable modality features by utilizing the structural and positional information. It consists of two main components: attended structure representation (ASR) and transformer-based part interaction (TPI). Specifically, ASR models the modality-invariant structure feature for each modality and dynamically selects the discriminative appearance regions under the guidance of the structure information. TPI mines the part-level appearance and position relations with a transformer to learn discriminative part-level modality features. With a weighted combination of ASR and TPI, the proposed SPOT explores the rich contextual and structural information, effectively reducing cross-modality difference and enhancing the robustness against misalignments. Extensive experiments indicate that SPOT is superior to the state-of-the-art methods on two cross-modal datasets. Notably, the Rank-1/mAP value on the SYSU-MM01 dataset has improved by 8.43%/6.80%. [ABSTRACT FROM AUTHOR]

As one of the fundamental tasks in scene understanding, 3D object detection from LiDAR point clouds has drawn extensive attention in the past few years. Although the existing voxel-based methods have achieved remarkable performance, how to effectively exploit geometric structure information of the point clouds to boost the detection performance remains to be explored. In this article, we propose a novel structure-information enhanced voxel network (SIEV-Net) for 3D object detection from LiDAR point clouds. The proposed SIEV-Net learns feature representations of 3D objects by jointly considering uneven spatial distribution and height information of the point clouds. Specifically, considering the uneven spatial distribution characteristics of point clouds, a hierarchical-voxel feature encoding module is proposed to effectively extract features of voxels in both sparse and dense regions. Besides, by utilizing the bird’s eye view (BEV) map of point clouds, a height information complement module is designed to minimize the height information lost in the process of point feature aggregation in a voxel network. Experimental results on the widely used KITTI benchmark dataset have demonstrated the efficacy of the proposed SIEV-Net. [ABSTRACT FROM AUTHOR]

In recent years, metal carbonate-based melts have been investigated as the promising heat transfer fluid and thermal energy storage medium for concentrating solar power plants. However, there are some limiting factors for investigating their structural information and properties. As the deep neural network develops, machine learning method is widely used in MD simulations. In this paper, interatomic potential driven by machine learning is developed based on datasets generated by first-principle molecular dynamics simulations in order to research the local structure and properties of molten Li2CO3-Na2CO3 binary salts at target temperatures. The machine learning potential enables higher efficiency and similar accuracy relative to DFT and yields precise descriptions of microstructures and properties. Microstructural evolution is analyzed through partial radial distribution functions. We observed lithium cation exhibits more coordination modes and the strength of Na+ with CO32− is weaker than that of Li+. Further, comparing the inter-ionic partial radial distribution function diagrams of binary melts and those of pure melt, interatomic characteristics were obtained. The evolution of properties calculated by machine learning, including density, self-diffusion coefficients, thermal conductivity, and viscosity over the entire operating temperature range, is documented. The relationships between properties and temperature or properties and the fraction of Li2CO3 were explored by the trained machine learning potential. This work exhibits a thorough understanding about the local structure and property of Li2CO3-Na2CO3 melt and reveals the accuracy of machine learning potential on molten binary carbonates for the first time. [ABSTRACT FROM AUTHOR]

Due to the lack of deep-level information mining and utilization in the existing network representation learning methods,the potential pattern structure similarity was proposed by further exploring the potential information in the network.The similarity score between network structures was defined to measure the similarity between various structures so that nodes could cross irrelevant vertices to obtain high-order similarities on the global structure.In order to achieve the best effect,deep learning was used to fuse multiple information sources to participate in training together to make up for the deficiency of random walks.In the experiment,Lap,DeepWalk,TADW,SDNE and CANE were selected as comparison methods,and three real-world networks were used as data sets to verify the validity of the model,and experiments of node classification and link reconstruction are carried out.In the node classification,the average performance is improved by 1.7 percentage points for different datasets and training proportions.In the link reconstruction experiment,only half the dimension is needed to achieve better performance.Finally,the performance improvement of the model under different network depths was discussed.By increasing the depth of the model,the average performance of node classification increased by 1.1 percentage points.

RNA-binding proteins play an important role in the biological process. However, the traditional experiment technology to predict RNA-binding residues is time-consuming and expensive, so the development of an effective computational approach can provide a strategy to solve this issue. In recent years, most of the computational approaches are constructed on protein sequence information, but the protein structure has not been considered. In this paper, we use a novel computational model of RNA-binding residues prediction, using protein sequence and structure information. Our hybrid features are encoded by local sequence and structure feature extraction models. Our predictor is built by employing the Granular Multiple Kernel Support Vector Machine with Repetitive Under-sampling (GMKSVM-RU). In order to evaluate our method, we use fivefold cross-validation on the RBP129, our method achieves better experimental performance with MCC of 0.3367 and accuracy of 88.84%. In order to further evaluate our model, an independent data set (RBP60) is employed, and our method achieves MCC of 0.3921 and accuracy of 87.52%. Above results demonstrate that integrating sequence and structure information is beneficial to improve the prediction ability of RNA-binding residues. [ABSTRACT FROM AUTHOR]

Due to the lack of deep-level information mining and utilization in the existing network representation learning methods, the potential pattern structure similarity was proposed by further exploring the potential information in the network. The similarity score between network structures was defined to measure the similarity between various structures so that nodes could cross irrelevant vertices to obtain high-order similarities on the global structure. In order to achieve the best effect, deep learning was used to fuse multiple information sources to participate in training together to make up for the deficiency of random walks. In the experiment, Lap, DeepWalk, TADW, SDNE and CANE were selected as comparison methods, and three real-world networks were used as data sets to verify the validity of the model, and experiments of node classification and link reconstruction are carried out. In the node classification, the average performance is improved by 1.7 percentage points for different datasets and training proportions. In the link reconstruction experiment, only half the dimension is needed to achieve better performance. Finally, the performance improvement of the model under different network depths was discussed. By increasing the depth of the model, the average performance of node classification increased by 1.1 percentage points. [ABSTRACT FROM AUTHOR]

• The structure and transport properties of molten Na 2 CO 3 -K 2 CO 3 salts are studied by machine learning molecular dynamics. • Deep potential models are trained on first-principles data and can accurately describe interatomic interactions. • The DP model achieves DFT-level precision, reproducing structural results from FPMD simulations. • Detailed structural information on the radial and angle distributions of ions is introduced. • The DP model is a more accurate predictor of properties than the empirical potential models. Molten Na 2 CO 3 -K 2 CO 3 binary salts have shown promise as electrolytes for fuel cell technology. Extensive investigations of the local structure and transport properties of these melts have been conducted via experiments and molecular dynamics (MD) simulations. With advancing machine learning capabilities, deep learning strategies are being widely incorporated into MD to generate highly accurate machine learning (ML) potentials. In this work, first-principle-derived Deep Potential is applied to study Na 2 CO 3 -K 2 CO 3 melts at specified temperatures. A deep neural network acts to fit the first-principle data and produce the model. This model enables large-scale MD simulations with accuracy comparable to density functional theory (DFT). The local structure is analyzed. Temperature and concentration impacts on ion arrangements are observed. The concentration effect has little influence on the melt structure, but the radial and angular distributions between the ions are strongly affected by temperature. Density, viscosity, and diffusion values predicted by the DP model agree closely with experimental measurements. In comparison to conventional empirical potentials, the DP model demonstrates DFT-level accuracy for structural analysis and experimental-level property predictions. This study provides detailed insights into the microstructure and transport properties of Na 2 CO 3 -K 2 CO 3 melts via the power of the DP model, showcasing their ability to advance understanding of complex molten salt systems. Meanwhile, a molten salt database is initially constructed using machine learning strategies. [ABSTRACT FROM AUTHOR]

Graph-based unsupervised feature selection methods have successfully processed high-dimensional data since they can effectively preserve data structure information. However, most existing methods are facing two challenges: (1) it is unreliable to exploit the intrinsic spatial structure from raw data because of the existence of redundant information and noise; (2) they only consider the first-order similar information but ignore the high-order similar information (i. e., samples with similar neighborhood network structure should be similar to each other). This study proposes an unsupervised feature selection with high-order similarity learning (HSL) to tackle the above problems. The projection matrix, first-order similar information, and high-order similar information are simultaneously learned in a unified framework, such that the intrinsic structure information can be exploited in the clean data that come from a latent low-dimensional embedding subspace. Specifically, we project raw data into a low-dimensional embedding subspace to effectively eliminate redundant information and noise. Moreover, we simultaneously consider the first-order similarity and the high-order similarity in the low-dimensional embedding space so that a better graph Laplacian matrix can be learned to fully preserve the intrinsic structure information of raw data and simultaneously utilize the label information. Furthermore, we design an effective optimization algorithm to tackle the proposed HSL. Comprehensive experiments on real-world datasets verify the superiority and effectiveness of HSL. [ABSTRACT FROM AUTHOR]

In this paper, we propose a method to enhance the performance of segmentation models for medical images. The method is based on convolutional neural networks that learn the global structure information, which corresponds to anatomical structures in medical images. Specifically, the proposed method is designed to learn the global boundary structures via an autoencoder and constrain a segmentation network through a loss function. In this manner, the segmentation model performs the prediction in the learned anatomical feature space. Unlike previous studies that considered anatomical priors by using a pre-trained autoencoder to train segmentation networks, we propose a single-stage approach in which the segmentation network and autoencoder are jointly learned. To verify the effectiveness of the proposed method, the segmentation performance is evaluated in terms of both the overlap and distance metrics on the lung area and spinal cord segmentation tasks. The experimental results demonstrate that the proposed method can enhance not only the segmentation performance but also the robustness against domain shifts.

Text classification is a foundational task in many natural language processing applications. All traditional text classifiers take words as the basic units and conduct the pre-training process (like word2vec) to directly generate word vectors at the first step. However, none of them have considered the information contained in word structure which is proved to be helpful for text classification. In this paper, we propose a word-building method based on neural network model that can decompose a Chinese word to a sequence of radicals and learn structure information from these radical level features which is a key difference from the existing models. Then, the convolutional neural network is applied to extract structure information of words from radical sequence to generate a word vector, and the long short-term memory is applied to generate the sentence vector for the prediction purpose. The experimental results show that our model outperforms other existing models on Chinese dataset. Our model is also applicable to English as well where an English word can be decomposed down to character level, which demonstrates the excellent generalisation ability of our model. The experimental results have proved that our model also outperforms others on English dataset. [ABSTRACT FROM AUTHOR]

When abstracting Chinese texts, the traditional TextRank algorithm only considers the similarity between nodes and neglects other important information of the text. Firstly, aiming at Chinese single document, on the basis of existing research, this paper used TextRank algorithm,on the one hand, it considered the similarities between sentences, on the other hand, Text-Rank was combined with the overall structural information of texts and the contextual information of sentences, such as the physical position of the document sentences or paragraph, feature sentences, core sentences and other sentences that might increase the weight of the sentence, all were used to generate the digest candidate sentence group of the text. And then, removing high-similarity sentences by redundancy processing technology on the digest candidate sentence group. Finally, the experimental verification shows that the algorithm can improve the accuracy of the generated digest, indicating the effectiveness of the algorithm. [ABSTRACT FROM AUTHOR]

Multi-view learning, such as multi-view feature learning and multi-view clustering, has a wide range of applications in machine learning and pattern recognition. Most previous studies employ the multiple data information from various views to improve the performance of learning. The key problem is to integrate the symbiotic part of the different views or datasets. In practical clustering task, the symbiotic part includes two levels: global structure information and local structure information. However, traditional multi-view clustering methods usually ignore the energy of the local structure information. This paper proposes a novel multi-view clustering model to solve this problem, which simultaneously integrates the global structure information and local structure information of all the views. By integrating the fusion of global spectral embedding and the fusion of spectral manifold embedding from multi-view data, we construct an objective function to find the final fusional embedding and give an iteration method to solve it by using the L2,1 norm. Finally, the K-means clustering method is applied to the obtained final fusional embedding. Extensive experimental results on several real multi-view data sets demonstrate the superior performance of our model. [ABSTRACT FROM AUTHOR]

Researchers have proposed a wide variety of visual attention models, ranging from models that use local, low-level image features to recent approaches that incorporate semantic information. However, most models do not account for the visual attention evident in images with certain global structures. We focus specifically on “leading line” structures, in which explicit or implicit lines converge at a point. Through this study, we have conducted the experiments to investigate the visual attentions in images with leading line structure and propose new models that combine the low-level feature of center-surround differences of visual stimuli, the semantic feature of center bias and the structure feature of leading lines. We also create a new data set from 110 natural images containing leading lines and the eye-tracking data for 16 subjects. Our evaluation experiment showed that our models outperform the existing models against common indicators of saliency-map evaluation, underscoring the importance of leading lines in the modeling of visual attention. [ABSTRACT FROM AUTHOR]

Multivariate finite mixture model is becoming more and more popular in image processing. Performing image denoising from image patches to the whole image has been widely studied and applied. However, there remains a problem that the structure information is always ignored when transforming the patch into the vector form. In this paper, we study the operator which extracts patches from image and then transforms them to the vector form. Then, we find that some pixels which should be continuous in the image patches are discontinuous in the vector. Due to the poor anti-noise and the loss of structure information, we propose a new operator which may keep more information when extracting image patches. We compare the new operator with the old one by performing image denoising in Expected Patch Log Likelihood (EPLL) method, and we obtain better results in both visual effect and the value of PSNR. [ABSTRACT FROM AUTHOR]

Detecting fine-grained changes among a set of images taken at different times is a challenging problem, which is important for the applications such as high-value scene monitoring and structural inspection. Existing methods use a geometry registration and lighting compensation pipeline to alleviate the ingredients that affect change detection. However, lighting compensation may introduce a lot of artifacts, e.g., inaccurate local lighting conditions and fine-grained change disappearance. The distortion introduced by general registration methods also makes the process of lighting correction more challenging. To solve these problems, in this paper, we propose a new phase-based change detection method, called as pb-FGCD, which first aligns two observation of different times and then extract structure information to handle lighting difference. Benefiting from the proposed weighted amplitude filter and iterative alignment strategy, we could obtain a better registration performance. The structure information extracts the lighting irrelevant components of an image and is more sensitive to structure changes compared with the lighting compensation based method, which is vital for detecting the changes with low intensity difference. Furthermore, we contribute a new change detection dataset, which contains 100 groups of the real-world data of the ancient Mogao murals and the Terra-Cotta. It is the first fine-grained change detection dataset focusing on the ancient mural. Our approach significantly improves the accuracy by more than 190% higher F1-measure than the state-of-the-art methods. • A novel fine-grained change detection framework. • A novel phase-based fine-grained registration method. • Finding structural information effective for fine-grained change detection. • Building a real-world dataset for fine-grained change detection. • Increasing the F1-measure by more than 190% over those state-of-the-art methods. [ABSTRACT FROM AUTHOR]

[Display omitted] • Introduce capsule network for multi-focus image fusion. • Utilize structure information to help locate focus regions. • Design structure-guided flow module to integrate structure features. Existing deep learning based methods have shown their advantages in multi-focus image fusion task. However, most methods still suffer from inaccurate focus region detection. In this paper, we employ the property of part-whole relationships embedded by the Capsule Network (CapsNet) to solve the problem. Specifically, we introduce CapsNet in multi-focus image fusion task, and design a structure-guided flow module, which fully utilizes structure information to help locate focus regions. CapsNet is introduced to extract structure features by supervising gradient information of the image. Compared with traditional convolutional neural networks (CNNs), CapsNet takes into account the correlation of features from different positions, such that it encodes more compact features. Once structure features are obtained, a flow alignment module is introduced to learn flow field between structure and image features, and propagate effectively structure features to image features to make confident focus region detection. Experimental results show the proposed method achieves robust fusion performance on three publicly available multi-focus datasets, and outperforms or is comparable to the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Graph neural networks (GNNs) are the dominant paradigm for modeling and handling graph structural data by learning universal node representation. The traditional way of training GNNs depends on a great many labeled data, which is time-consuming and money-consuming. In some special scenes, it is even unavailable and impracticable. Self-supervised representation learning, which can generate labels by graph structural data itself, is a potential approach to tackle this problem. And turning to research self-supervised learning problems for heterogeneous graphs is more challenging than dealing with homogeneous graphs, there are fewer studies about it as well. In this paper, we propose a SE lf-supervised learning method for heterogeneous graph via S tructure I nformation based on M etapath (SESIM). Firstly, the pseudo-labels are constructed to train pretext tasks, using data itself and avoiding time-consuming manual labeling. Afterward, we use traditional graph neural networks to aggregate node features, obtaining the node embeddings. And then, the primary task and pretext tasks are designed by these node embeddings. The pretext tasks, i.e., jump numbers prediction between nodes in each metapath, can improve the representation ability of the primary task. Moreover, predicting jump numbers in each metapath can effectively utilize graph structural information, which is the essential property of nodes. Therefore, SESIM deepens the understanding of models for graph structure. At last, we train the primary task and pretext tasks jointly and balance the contributions of pretext tasks for the primary task. The key advantage of our proposed model is that we research self-supervised learning for the heterogeneous graph to address the time-consuming and money-consuming problem of obtaining labels. And we design a novel pretext task, i.e., jump numbers prediction in each metapath, via graph structural information based on the metapath. Empirical results validate the performance of the SESIM method and demonstrate that this method can improve the representation ability of traditional neural networks on link prediction tasks and node classification tasks. • The self-supervised information is generated by jump number prediction in each metapath which bases on the graph structure. • We apply self-supervised learning based on metapath on heterogeneous graphs. • The experimental results show that the pretext tasks can improve the performance of primary task and representation power of GNNs. [ABSTRACT FROM AUTHOR]

In high-resolution synthetic aperture radar (SAR) images, targets are often spatially spread, and some separated regions of one target may be detected as different potential targets. The target should be represented by a part model to combine the separated regions into one target. In this letter, a part often means a separated region, and the part model obtains a hierarchical combination of the other parts. Moreover, the bottom hierarchy consists of strong scatters in the local SAR image, which are searched based on compressed intensities. The middle hierarchy consists of parts, which are generated based on connectivity and similarity of scatters in the bottom hierarchy. The top hierarchy delineates the relationship of parts in the middle hierarchy. A search method is designed based on the different features of the elements in each hierarchy. Consequently, the proposed algorithm combines different traits of targets to improve detection, and unites the separated regions of one target in the top hierarchy. This algorithm is validated by SAR images with different resolutions and scenes. [ABSTRACT FROM PUBLISHER]

Accumulated studies have shown that environmental factors (EFs) can regulate the expression of microRNA (miRNA) which is closely associated with several diseases. Therefore, identifying miRNA-EF associations can facilitate the study of diseases. Recently, several computational methods have been proposed to explore miRNA-EF interactions. In this paper, a novel computational method, MEI-BRWMLL, is proposed to uncover the relationship between miRNA and EF. The similarities of miRNA-miRNA are calculated by using miRNA sequence, miRNA-EF interaction, and the similarities of EF-EF are calculated based on the anatomical therapeutic chemical information, chemical structure and miRNA-EF interaction. The similarity network fusion is used to fuse the similarity between miRNA and the similarity between EF, respectively. Further, the multiple-label learning and bi-random walk are employed to identify the association between miRNA and EF. The experimental results show that our method outperforms the state-of-the-art algorithms.

This paper presents a novel depth video estimation method, which improves estimation property and refines the temporal consistency and spatial accuracy at the same time. The main idea is to incorporate more useful motion and structure information into depth estimation. First, an adaptive weight is calculated based on the motion information of adjacent frames and attached to a temporal term to update the energy function, thus reducing the matching errors and improving temporal consistency. Second, the depth continuity/discontinuity in spatial domain is properly judged by combining the edges of initial depth map and color segmentation, and used in the refinement strategy. Finally, we evaluate the performance of our algorithm with several public stereoscopic video sequences. Experimental results show that the qualities of depth video and synthesized virtual view are all improved. [ABSTRACT FROM PUBLISHER]