Fabric image retrieval is a meaningful issue, due to its potential values in many areas such as textile product design, e-commerce, and inventory management. Meanwhile, it is challenging because of the diversity of fabric appearance. Encourage by the recent breakthrough in the deep convolutional neural network (CNN), a deep learning framework is applied for fabric image retrieval. The idea of the proposed framework is that the binary code and feature for representing the image can be learning by a deep CNN when the data labels are available. The proposed framework employs a hierarchical search strategy that includes coarse-level retrieval and fine-level retrieval. Otherwise, a large-scale wool fabric image retrieval dataset named WFID with about 20 000 images are built to validate the proposed framework. The longitudinal comparison experiments for self-parameter optimization and horizontal comparison experiments for verifying the superiority of the algorithm are performed on this data set. The comparison experimental results indicate the superiority of the proposed framework.

/pdf/fabric-image-retrieval-system-using-hierarchical-search-fp88vmjaua.pdf

Fabric Image Retrieval System Using Hierarchical Search Based on Deep Convolutional Neural Network

Learning non-metric visual similarity for image retrieval

In this paper, we devise an efficient dissipation-preserving fourth-order difference solver for the fractional-in-space nonlinear wave equations. First of all, we present a detailed derivation of the discrete energy dissipation property of the system. Then, with the help of the mathematical induction and Brouwer fixed point theorem, it is shown that the proposed scheme is uniquely solvable. Subsequently, by virtue of utilizing the discrete energy method, it is proven that the proposed solver achieves the convergence rates of $${\mathcal {O}}(\Delta t^2+h^{4})$$
 in the discrete $$L^{\infty }$$
 - norm, and is unconditionally stable. And moreover, the exhibited convergence analysis is unconditional for the time step and space size, in comparison with the restrictive conditions required in the existing works. Finally, numerical results confirm the efficiency of the proposed scheme and exhibit the correctness of theoretical results.

An Effective Dissipation-Preserving Fourth-Order Difference Solver for Fractional-in-Space Nonlinear Wave Equations

In this paper, we develop and analyze the energy conservative time high-order AVF compact finite difference methods for variable coefficient acoustic wave equations in two dimensions. We first derive out an infinite-dimensional Hamiltonian system for the variable coefficient wave equations and apply the spatial fourth-order compact finite difference operator to the equations of the system to obtain a semi-discrete approximation system, which can be cast into a canonical finite-dimensional Hamiltonian form. We then apply the second-order and fourth-order AVF techniques to propose the fully discrete energy conservative time high-order AVF compact finite difference methods for wave equations in two dimensions. We prove that the proposed semi-discrete and fully-discrete schemes satisfy energy conservations in the discrete forms. We further prove that the semi-discrete scheme has the fourth-order convergence order in space and the fully-discrete AVF compact finite difference method has the fourth-order convergence order in both time and space. Numerical tests confirm the theoretical results.

The Conservative Time High-Order AVF Compact Finite Difference Schemes for Two-Dimensional Variable Coefficient Acoustic Wave Equations

Color is difficult to distinguish by human vision and is described by keywords, resulting in low efficiency of wool fabric retrieval in factories at present. To obtain the process sheets of existin...

Image retrieval of wool fabric. Part I: Based on low-level texture features:

Learning Deep Similarity Models with Focus Ranking for Fabric Image Retrieval

A new geometric modeling approach for woven fabric based on Frenet frame and Spiral Equation

Machine learning and deep learning have facilitated various successful studies of molecular property predictions. The rapid development of natural language processing and graph neural network (GNN) further pushed the state-of-the-art prediction performance of molecular property to a new level. A geometric graph could describe a molecular structure with atoms as the nodes and bonds as the edges. Therefore, a graph neural network may be trained to better represent a molecular structure. The existing GNNs assumed homogeneous types of atoms and bonds, which may miss important information between different types of atoms or bonds. This study represented a molecule using a heterogeneous graph neural network (MolHGT), in which there were different types of nodes and different types of edges. A transformer reading function of virtual nodes was proposed to aggregate all the nodes, and a molecule graph may be represented from the hidden states of the virtual nodes. This proof-of-principle study demonstrated that the proposed MolHGT network improved the existing studies of molecular property predictions. The source code and the training/validation/test splitting details are available at https://github.com/zhangruochi/Mol-HGT.

https://pubs.acs.org/doi/pdf/10.1021/acsomega.1c06389

Daiguo Deng

Papers

Learning Deep Similarity Models with Focus Ranking for Fabric Image Retrieval

A new geometric modeling approach for woven fabric based on Frenet frame and Spiral Equation

Describe Molecules by a Heterogeneous Graph Neural Network with Transformer-like Attention for Supervised Property Predictions