A5100706323- Advanced Vision and Imaging
- Computer Graphics and Visualization Techniques
- Advanced Image Processing Techniques
- Advanced Image and Video Retrieval Techniques
- Video Surveillance and Tracking Methods
- Image and Object Detection Techniques
- Topic Modeling
- Advanced Neural Network Applications
- Natural Language Processing Techniques
- Remote-Sensing Image Classification
- Pulsars and Gravitational Waves Research
- Image and Signal Denoising Methods
- Infrared Target Detection Methodologies
- Silicon and Solar Cell Technologies
- Advanced Measurement and Detection Methods
- Software Engineering Research
- Radio Astronomy Observations and Technology
- Image Processing Techniques and Applications
- Image Retrieval and Classification Techniques
- Semiconductor materials and interfaces
- Robotics and Sensor-Based Localization
- Landslides and related hazards
- Generative Adversarial Networks and Image Synthesis
- Smart Agriculture and AI
- Adaptive optics and wavefront sensing
Beijing Normal University
2015-2024
Xi'an Shiyou University
2024
Inner Mongolia University
2023-2024
Fudan University
2023
Yunnan University
2023
Beijing Institute of Big Data Research
2023
North China University of Technology
2023
Beijing Microelectronics Technology Institute
2023
Chang'an University
2023
Beijing University of Technology
2006-2022
Image segmentation is an important preprocessing operation in image recognition and computer vision. This paper proposes adaptive K-means method, which generates accurate results with simple avoids the interactive input of K value. method transforms color space images into LAB firstly. And value luminance components set to a particular value, order reduce effect light on segmentation. Then, equivalent relation between values number connected domains after setting threshold used segment...
Abstract Quickly and conveniently identifying buildings in disaster areas plays an important role assessment. To achieve the technical requirements of flood relief projects, this paper proposes a building extraction method for use with remote sensing images that combines traditional digital image processing methods convolution neural networks. First, threshold segmentation is used to select construct training dataset. Then, variety preprocessing are enhance selected Finally, improved Mask...
The tunable bandgaps and facile fabrication of metal halide perovskites make them attractive for tandem solar cells. One the main bottlenecks to achieve high-performance stable perovskite-based tandems is...
To estimate the parameters of fuzzy linear regression model with crisp input and output, we construct a method based on minimizing least square errors, propose some relative conclusions including normal equations, minimum solution, unique solution their analytical expressions parameters. Finally, one numerical example is used to illustrate our proposed methods reasonable.
In this paper, the passivation quality of crystalline silicon (c-Si) wafers, when passivated by atomic layer deposited aluminum oxide (ALD AlO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</sub> ), is investigated. Specifically, we investigated effect surface modification c-Si interface prior to ALD deposition (via -H and -OH termination wafer) over a large range thicknesses (0.4-80 nm). Fourier transform infrared (FTIR) studies confirmed...
Abstract This paper investigates a multi-resolution digital Earth model called PYXIS, which was developed by PYXIS Innovation Inc. The hexagonal grids employ an efficient hierarchical labeling scheme for addressing pixels. We provide recursive definition of the grids, systematic approach to labeling, algorithm add labels, and discussion discrete Fourier transform on grids.
High-resolution digital elevation models (DEMs) are important for relevant geoscience research and practical applications. Compared with traditional hardware-based methods, super-resolution (SR) reconstruction techniques currently low-cost feasible methods used obtaining high-resolution DEMs. Single-image (SISR) have become popular in DEM SR recent years. However, has not yet utilized reference-based image (RefSR) techniques. In this paper, we propose a terrain self-similarity-based...
Models designed to detect abnormalities that reflect disease from facial structures are an emerging area of research for automated analysis, which has important potential value in smart healthcare applications. However, most the proposed models directly analyze whole face image containing background information, and rarely consider effects different regions on analysis results. Therefore, view these effects, we propose end-to-end attention network with spatial transformation estimate pain...
A new model called the Transformer-Unet Generative Adversarial Network (TUGAN) is proposed for super-resolution reconstruction of digital elevation models (DEMs). Digital are used in many fields, including environmental science, geology and agriculture. The uses a self-similarity Transformer (SSTrans) as generator U-Net discriminator. SSTrans, that we previously proposed, can yield good results structurally complex areas but has little advantage when surface simple smooth because too...
The prior work on natural language inference (NLI) debiasing mainly targets at one or few known biases while not necessarily making the models more robust. In this paper, we focus model-agnostic strategies and explore how to (or is it possible to) make NLI robust multiple distinct adversarial attacks keeping even strengthening models’ generalization power. We firstly benchmark prevailing neural including pretrained ones various datasets. then try combat by modifying a mixture of experts...
Precipitation nowcasting has long been a challenging problem in meteorology. While recent studies have introduced deep neural networks into this area and achieved promising results, these models still struggle with the rapid evolution of rainfall extremely imbalanced data distribution, resulting poor forecasting performance for convective scenarios. In article, we evaluate amount information different precipitation tasks varying lengths using mutual information. We propose two strategies:...
In application scenarios such as UAV inspection, deep learning-based object detection methods are increasingly used to improve the automation of line inspection. aerial view scene, drone is usually fly at a high altitude from ground, so proportion in image relatively small. When YoloV3 network identifies small objects, result would not be good because there less information 8x downsampling feature map. this paper, base on LaSOT data set, has been modified by adjusting values anchors and...
Abstract Pulsar research has been a hot topic in the area of astronomy since they were first discovered. discovery is fundamental for pulsar research. While pulsars are now visible across electromagnetic spectrum, searches with modern radio telescopes most promising. As performance astronomical instruments improves, number candidates detected by grows at an exponential rate. The application artificial intelligence to field pulsar-candidate identification can automatically and efficiently...
<abstract> <p>Estimating the volume of food plays an important role in diet monitoring. However, it is difficult to perform this estimation automatically and accurately. A new method based on multi-layer superpixel technique proposed paper avoid tedious human-computer interaction improve accuracy. Our includes following steps: 1) obtain a pair images along with depth information using stereo camera; 2) reconstruct plate plane from disparity map; 3) warp input image map form...
The rapid development of web services has made it increasingly challenging for developers to find desired services. To address this issue, researchers have developed various powerful models service recommender systems. Recently, graph neural networks shown promising performance in deep learning tasks including recommendation. This paper proposes a novel network recommendation using hierarchical attention mechanism that combines node-level and motif-level mechanisms. is responsible...
Abstract In clinical medicine, the pain feeling is a significant indicator for medical condition of patients. Of late, automatic assessment methods have received more and interests. Many researchers proposed corresponding achieved impressive results. However, they always ignore locality individual differences painful expression. Therefore, identity aware network (LIAN) presented here. Concretely, characteristic, module consisting two‐branch structure, feature attention branches, presented....
The software systems which are related to national projects always very crucial. This kind of involves hi-tech factors and has spend a large amount money, so the quality reliability deserve be further studied. Hence, we propose apply three classification techniques most used in data mining fields: Bayesian belief networks (BBN), nearest neighbor (NN) decision tree (DT), validate usefulness metrics for risk prediction. Results show that comparing with such as Lines code (LOQ Cyclomatic...
Many recent studies have shown that for models trained on datasets natural language inference (NLI), it is possible to make correct predictions by merely looking at the hypothesis while completely ignoring premise. In this work, we manage derive adversarial examples in terms of hypothesis-only bias and explore eligible ways mitigate such bias. Specifically, extract various phrases from hypotheses (artificial patterns) training sets, show they been strong indicators specific labels. We then...
In the research of using Radial Basis Function Neural Network (RBF NN) forecasting nonlinear time series, we investigate how different clusterings affect process learning and forecasting. We find that κ-means clustering is very suitable. order to increase precision introduce a feedback term escape from local minima energy, then use model forecast series which are produced by Mackey–Glass equation stocks. By selecting suitable term, much better results obtained.
In the traditional multi-resolution texture mapping method for large-scale terrain rendering, resolution of is only decided by distance to viewpoint. some sharp region stretched a lot and brings distortion problems, which affect visual effect virtual scene. order solve this problem, many methods were proposed. The most universal gain with higher map it onto regions. But in finest level has no available. By analyzing nature attribute slope, we propose an efficient upsampling adaptive based on...