A5101578854- Remote-Sensing Image Classification
- Chaos-based Image/Signal Encryption
- Advanced Image and Video Retrieval Techniques
- Handwritten Text Recognition Techniques
- Chaos control and synchronization
- CO2 Reduction Techniques and Catalysts
- Natural Language Processing Techniques
- Vehicle License Plate Recognition
- Traditional Chinese Medicine Studies
- Advanced Steganography and Watermarking Techniques
- Generative Adversarial Networks and Image Synthesis
- Quantum chaos and dynamical systems
- Simulation and Modeling Applications
- Electrocatalysts for Energy Conversion
- Infrared Target Detection Methodologies
- Mathematical Dynamics and Fractals
- Image Enhancement Techniques
- Web Data Mining and Analysis
- Advanced Differential Equations and Dynamical Systems
- Ionic liquids properties and applications
- Advanced Neural Network Applications
- Face recognition and analysis
- Image Processing and 3D Reconstruction
- Advanced Image Fusion Techniques
- Medical Research and Treatments
Inner Mongolia Electric Power Survey & Design Institute (China)
2024-2025
Zhengzhou Institute of Emerging Industrial Technology
2024-2025
East China University of Science and Technology
2023-2025
State Key Laboratory of Chemical Engineering
2023-2025
Nanjing University of Finance and Economics
2024
Institute of Process Engineering
2024
Chongqing Institute of Green and Intelligent Technology
2013-2023
Nantong University
2008-2021
Heilongjiang University of Science and Technology
2021
University of Chinese Academy of Sciences
2019-2021
In this paper we address the problem of automatically parsing fashion images with weak supervision from user-generated color-category tags such as “red jeans” and “white T-shirt”. This is very challenging due to large diversity items absence pixel-level tags, which make traditional fully supervised algorithms inapplicable. To solve problem, propose combine human pose estimation module, MRF-based color category inference module (super)pixel-level classifier learning generate multiple...
Abstract The performance of Lithium–sulfur (Li–S) batteries is constrained by the migration lithium polysulfide (LiPS), slow conversion LiPS, and significant reaction barrier encountered during precipitation/dissolution Li 2 S throughout discharge/charge cycle. In this contribution, study presents Ni–Co dual‐atom catalytic sites on hollow nitrogen‐doped carbon (NiCoNC). Theoretical calculations experimental data reveal that catalysts (DACs) accelerate kinetic LiPSs facilitate...
Quantifying the dynamic changes and relationships between ecosystem services (ESs) land use change is critical for sustainable management optimization. However, comprehensive discussions on spatiotemporal variations of ESs their with intensity (LUI) are lacking, particularly in context significant climate warming. Systematic analyses forestry unit scale limited, leading to considerable uncertainty management, especially alpine ecosystems Northern Hemisphere, where have significantly...
Beauty e-Experts, a fully automatic system for makeover recommendation and synthesis, is developed in this work. The synthesis simultaneously considers many kinds of items on hairstyle makeup. Given user-provided frontal face image with short/bound hair no/light makeup, the e-Experts not only recommends most suitable hairdo but also synthesizes virtual makeup effects. To acquire enough knowledge beauty modeling, we built Database, which contains 1,505 female photos variety attributes...
The practical application of sulfur (S) cathodes in lithium‐sulfur (Li‐S) batteries is hindered by the shuttling soluble lithium polysulfides (LiPSs) and sluggish redox kinetics. Addressing these challenges requires advanced catalytic host materials capable trapping LiPSs accelerating Li‐S reactions. However, single‐site catalysts struggle to effectively mediate complex multi‐step multi‐phase conversion processes. In this study, we present a novel dual‐site catalyst, featuring nickel single...
Beauty e-Experts, a fully automatic system for hairstyle and facial makeup recommendation synthesis, is developed in this work. Given user-provided frontal face image with short/bound hair no/light makeup, the e-Experts can not only recommend most suitable hairdo but also show synthetic effects. To obtain enough knowledge beauty modeling, we build Database, which contains 1,505 attractive female photos variety of attributes beauty-related annotated. Based on Dataset, two problems are...
Automatic License plate recognition (ALPR) remains a challenging task in face of some difficulties such as multi-line character distribution and license (LP) deformation due to camera angles. Most existing ALPR methods either focus on single-line LP or perform horizontal detection with character-level annotations. In this paper, we propose novel end-to-end irregular (EILPR) detect recognize the text arbitrary shooting angles, using only plate-level annotations for training. EILPR,...
Adding Ni nanoparticles (NPs) is an effective methodology to improve the catalytic activity of single-atom catalysts (SACs) in electrochemical CO2 reduction reaction (CO2RR). However, effect form which NPs exist on SAs needs further exploration. Herein, interaction between different forms existence and investigated g-C3N4-inducing carbon nanosheets. Comparing with impregnated NPs, situ-doped are encapsulated by a graphitic layer avoid catalyzing hydrogen evolution (HER) facilitate adsorption...
Hashing-based methods seek compact and efficient binary codes that preserve the similarity between data. For most existing hashing methods, an input (e.g. image) is first encoded as a vector of hand-crafted visual feature, followed by hash projection quantization step to obtain vector. Most features only encode low-level information input, feature may not semantic similarities pairwise inputs. Meanwhile, function learning process independent with representation, so be optimal for projection....
This paper discusses about a fusion algorithm for infrared and visible light images based on region segmentation the dual-tree complex wavelet transform. Before image segmentation, morphological top-hat filtering is firstly performed IR visual respectively details of luminous area are eliminated. Morphological bottom-hat then two kinds dark Make filtered subtract obtain enhanced images. Then threshold method used to segment After Dual-Tree Complex Wavelet Transform (DTCWT) coefficients from...
A finite set statistics (FISST)-based method is proposed for multi-target tracking in the image plane of optical sensors. The involves using signal amplitude information probability hypothesis density (PHD) filter which derived from FISST to improve performance. signals generated by sensor modeled first, likelihood ratio between target and clutter derived. An alternative approach adopted situations where noise (SNR) unknown. Then PHD recursion equations incorporated with are Gaussian mixture...
License plate recognition (LPR) remains a challenging task in face of some difficulties such as image deformation and multi-line character distribution. Text rectification that is crucial to eliminate the effects has attracted increasing attentions scene text recognition. However, current methods are not designed specifically for LPR, which did take features into account. Considering fact license (LP) can only generate perspective distortion due its rigid feature, this paper we propose novel...
The existing object detection algorithms have low recognition accuracy for prohibited items due to the complex background, large variation of target scale, and mutual occlusion objects in X-ray security inspection images. In order accurately identify real-time, an image algorithm based on improved YOLOv4 is proposed. Firstly, deformable convolution introduced into network improve feature extraction ability items. Then, GHM loss used optimize function, so that model can focus difficult...
In order to protect color digital image's information, a new scrambling algorithm based on chaotic sequence is proposed. Firstly, according the key and size of image m×n, with length 3mn generated, corresponding subscript generated by sorting it. Then, RGB component combined matrix pixels, transforming pixel coordinates will be realized. Lastly, pixels decomposed into component, generated. The effect security also analyzed using evaluation index. Theoretical analysis experimental results...
Scene text, especially irregular is difficult to recognize due the arbitrary-oriented characters and arrangement. Most existing methods address text by rectifying it into a regular one, which achieve good performance. However, these are possible remove character information in some curved texts. To overcome this issue, we focus on extracting features that robust orientation changes instead of rectifying. In work, propose an end-to-end trainable model combines Gabor Convolutional Network...