A5100603785- Conservation Techniques and Studies
- Cultural Heritage Materials Analysis
- Building materials and conservation
- Geological Modeling and Analysis
- Image and Signal Denoising Methods
- Advanced Image Fusion Techniques
- Hydrocarbon exploration and reservoir analysis
- Geoscience and Mining Technology
- Geomechanics and Mining Engineering
- Direction-of-Arrival Estimation Techniques
- Power Systems Fault Detection
- HVDC Systems and Fault Protection
- Advanced Vision and Imaging
- Islanding Detection in Power Systems
- Geological and Geophysical Studies
- Speech and Audio Processing
- Soil, Finite Element Methods
- Blind Source Separation Techniques
- 3D Surveying and Cultural Heritage
- Advanced SAR Imaging Techniques
- Drilling and Well Engineering
- Optical measurement and interference techniques
- Remote Sensing and Land Use
- Handwritten Text Recognition Techniques
- Dam Engineering and Safety
University of Science and Technology Beijing
2020-2025
Guilin University of Electronic Technology
2024
National University of Defense Technology
2007-2023
Ningbo University
2007-2021
Amazon (United States)
2020-2021
Zhejiang University
2013-2020
Sun Yat-sen University
2005-2019
Xi'an Jiaotong University
2017-2018
Hohai University
2016
Tongji University
2016
Single image rain streaks removal has recently witnessed substantial progress due to the development of deep convolutional neural networks. However, existing learning based methods either focus on entrance and exit network by decomposing input into high low frequency information employing residual reduce mapping range, or introduction cascaded scheme decompose task multi-stages. These treat as an encapsulated end-to-end module without deepening rationality superiority design. In this paper,...
Condition-based maintenance is critical to reduce the costs of and improve production efficiency. Data-driven method based on neural network (NN) one most used models for mechanical components condition recognition. In this paper, we introduce a new bearing recognition multifeatures extraction deep (DNN). First, calculates time domain, frequency time-frequency domain features represent characteristic vibration signals. Then nonlinear dimension reduction algorithm learning proposed redundancy...
Recent progress in biomedical image segmentation based on deep convolutional neural networks (CNNs) has drawn much attention. However, its vulnerability towards adversarial samples cannot be overlooked. This paper is the first one that discovers all CNN-based state-of-the-art models are sensitive to perturbations. limits deployment of these methods safety-critical fields. In this paper, we discover global spatial dependencies and contextual information a can exploited defend against attacks....
Abstract Craquelure is the most common defect on ancient polychrome paintings, which may deteriorate further to paint loss. Previous image processing methods, can accurately recognize loss, have limited precision and efficiency in segmenting craquelure. This paper proposes a semantic segmentation method, Res-UNet, for recognition of craquelure loss Palace Museum, Beijing. The residual structure ResNet-50 enables avoidance network degradation, features be fully extracted. Using unique skip...
Cultural heritage objects, including traditional Chinese polychrome paintings on architectures (Caihua) and wooden architectural components, frequently exhibit surface defects that are highly sensitive to environmental factors, resulting in progressive deterioration. However, due limited data acquisition methods quantitative analysis models, the stability risks of such as cracks during changes remain unclear. This study integrates photogrammetry digital image processing investigate...
Abstract In the past few decades, a series of research, restoration and protection have been performed on ancient murals in Mogao Grottoes. These treatments played significant roles conservation elegant art Dunhuang. But whether reinforcement are still active which materials more appropriate remain unknown. issues crucial for murals. On-site investigation Grottoes has carried out. The purpose this is to examine renewed diseases repaired before. It will also provide useful information about...
Understanding product attributes plays an important role in improving online shopping experience for customers and serves as integral part constructing a knowledge graph. Most existing methods focus on attribute extraction from text description or utilize visual information images such shape color. Compared to the inputs considered prior works, image fact contains more information, represented by rich mixture of words clues with layout carefully designed impress customers. This work proposes...
Can one build a knowledge graph (KG) for all products in the world? Knowledge graphs have firmly established themselves as valuable sources of information search and question answering, it is natural to wonder if KG can contain about offered at online retail sites. There been several successful examples generic KGs, but organizing poses many additional challenges, including sparsity noise structured data products, complexity domain with millions product types thousands attributes,...
Single image rain streaks removal has recently witnessed substantial progress due to the development of deep convolutional neural networks. However, existing learning based methods either focus on entrance and exit network by decomposing input into high low frequency information employing residual reduce mapping range, or introduction cascaded scheme decompose task multi-stages. These treat as an encapsulated end-to-end module without deepening rationality superiority design. In this paper,...
Abstract In 2013, several tombs were discovered and excavated in Southeastern Beijing, China, yielding various burial products. Among these pottery towers, a representative artifact of the Han Dynasty rarely found area. Many studies on architectural aspects, such as construction type situation, have been conducted based but only few examined their raw materials or pigments. this study, black, white, red pigments identified carbon calcite, cinnabar, respectively, through multi-analysis...
Dew is a common but important phenomenon. Though water previously considered to be threat earthen heritage sites, artificial dew showing potential in relic preservation. A model of prediction on sites will essential for developing preventive protection methods, studies formation processes relics are limited. In this study, two parameter proposed. It makes approximations according the features assuming that thin and steady air layer exists close air–solid interface. This semi-theoretical was...
The inherent speckle noise in synthetic aperture radar (SAR) images severely degrades the image interpretation and affects follow-up image-processing tasks. Thus, suppression is a critical step SAR preprocessing. We propose novel locally adaptive reduction algorithm based on Bayesian maximum posteriori (MAP) estimation wavelet domain. First, presented method performs logarithmical transform to original speckled an undecimated (UWT). Rayleigh distribution used model statistics of...
This paper presents scientific analyses of the wall paintings in Wudang Lamasery, which is located Baotou area Inner Mongolia, China. Raman spectroscopy was used to analyze pigments mural Lam rim Hall. The results show that vermilion, red lead, chrome yellow, emerald green and synthetic ultramarine were used. existence provides a clue for date painted.
A digital watermarking algorithm is proposed basing on the stationary wavelet transform to embed watermark in primitive DEM images and withdraw from watermarked images. Considering characteristic of images, embedding position selected relatively gentle terrains by analyzing slope image. The intensity optimized an optimization model, which objective function defined as similarity between original water mark abstracted watermark. model then resolved with ant colony algorithm, precision data...
As the structural information of an object can be well descripted by edge pixels, we observe that greatest challenge for locating text edges on scene image is how to handle edge-adhesion problem. In this paper propose Skeleton-cut Text Detector, which take text-specific cues such as a novel presentation skeleton into account hunt efficiently with improved recall rate. To address problem, skeleton-junctions detection and elimination are performed first cut candidate out map. Then candidates...
Recent progress in biomedical image segmentation based on deep convolutional neural networks (CNNs) has drawn much attention. However, its vulnerability towards adversarial samples cannot be overlooked. This paper is the first one that discovers all CNN-based state-of-the-art models are sensitive to perturbations. limits deployment of these methods safety-critical fields. In this paper, we discover global spatial dependencies and contextual information a can exploited defend against attacks....