- Domain Adaptation and Few-Shot Learning
- Advanced Image and Video Retrieval Techniques
- Multimodal Machine Learning Applications
- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- COVID-19 diagnosis using AI
- Anomaly Detection Techniques and Applications
- Natural Language Processing Techniques
- Human Pose and Action Recognition
- Face and Expression Recognition
- Remote-Sensing Image Classification
- Topic Modeling
- Image Processing Techniques and Applications
- Image Retrieval and Classification Techniques
- Advanced Image Processing Techniques
- Radiomics and Machine Learning in Medical Imaging
- Physics of Superconductivity and Magnetism
- Advanced Vision and Imaging
- Remote Sensing and Land Use
- Advanced Image Fusion Techniques
- Video Analysis and Summarization
- Adversarial Robustness in Machine Learning
- Geophysical Methods and Applications
- Manufacturing Process and Optimization
- Cryptography and Data Security
Renmin University of China
2024-2025
University of Shanghai for Science and Technology
2025
Wuhan University of Technology
2010-2024
Harbin Engineering University
2024
Sanya University
2024
National University of Defense Technology
2023-2024
Bank of China
2024
Beihang University
2012-2023
Shanghai University of Engineering Science
2011-2023
Shantou University
2023
Zero-shot learning (ZSL) aims to recognize novel classes by transferring semantic knowledge from seen unseen classes. Since is built on attributes shared between different classes, which are highly local, strong prior for localization of object attribute beneficial visual-semantic embedding. Interestingly, when recognizing images, human would also automatically gaze at regions with certain clue. Therefore, we introduce a goal-oriented estimation module (GEM) improve the discriminative based...
Masked Autoencoder (MAE) has recently been shown to be effective in pre-training Vision Transformers (ViT) for natural image analysis. By reconstructing full images from partially masked inputs, a ViT encoder aggregates contextual information infer regions. We believe that this context aggregation ability is particularly essential the medical domain where each anatomical structure functionally and mechanically connected other structures Because there no ImageNet-scale dataset pre-training,...
Compared with color or grayscale images, hyperspectral images deliver more informative representation of ground objects and enhance the performance many recognition classification applications. However, are normally corrupted by various types noises, which have a serious impact on subsequent image processing tasks. In this paper, we propose novel denoising method based tucker decomposition to model nonlocal similarity across spatial domain global along spectral domain. method, 3-D full band...
Subspace clustering is a useful technique for many computer vision applications in which the intrinsic dimension of high-dimensional data smaller than ambient dimension. Traditional subspace methods often rely on self-expressiveness property, has proven effective linear clustering. However, they perform unsatisfactorily real with complex nonlinear subspaces. More recently, deep autoencoder based have achieved success owning to more powerful representation extracted by network. Unfortunately,...
Abstract Carbon fiber‐reinforced composites are widely used in aerospace, rail transportation, and new energy vehicles due to their small specific gravity as well good mechanical chemical properties. Additive manufacturing of continuous carbon polymer is an innovative composite technique. In this paper, bending shear specimens with different hatch spacing layer thicknesses were prepared by fused filament fabrication (FFF) technique subjected three‐point experiments interlayer experiments,...
While the research on convolutional neural networks (CNNs) is progressing quickly, real-world deployment of these models often limited by computing resources and memory constraints. In this paper, we address issue proposing a novel filter pruning method to compress accelerate CNNs. Our work based linear relationship identified in different feature map subspaces via visualization maps. Such implies that information CNNs redundant. eliminates redundancy filters applying subspace clustering...
Hyperspectral unmixing is a crucial task for hyperspectral images (HSIs) processing, which estimates the proportions of constituent materials mixed pixel. Usually, pixels can be approximated using linear mixing model. Since each material only occurs in few real HSI, sparse nonnegative matrix factorization (NMF), and its extensions are widely used as solutions. Some recent works assume that distributed certain structures, added constraints to NMF However, they consider spatial distribution...
The protection of agricultural heritage sites has become a global human responsibility and consensus. However, the potential effect on green development agriculture currently been ignored. Since ancient times, China founded agriculture, number important cultural heritages ranks first in world, with strong representativeness. two-way fixed effects model was employed to empirically test positive impact site utilising data from 30 provincial units over 21-year period 2001 2021 this paper....
The digital economy offers new solutions for reconciling the growth of non-timber forest-based (NTFBE) with ecological and environmental protection. Utilizing panel data from China’s provinces between 2011 2020, this study constructed a comprehensive indicator system purpose examining coordinated development NTFBE environment. employment econometric methods, including Tobit models, mediated effects spatial Durbin models threshold regression has enabled us to ascertain that can effectively...
Background Caregivers of patients with dementia or Alzheimer's disease (AD) face special health challenges due to the progressive nature disease. Self‐care has crucial importance on individuals’ management life, health, and well‐being. However, limited evidence is available self‐care AD caregivers. This study aimed investigate influence caregiving caregivers based caregivers’ experience. In addition, facilitators were assessed. Methods A sample 45 was recruited from a local community in...
Parkinson's disease (PD) is characterized by microglia activation that leads to neuroinflammation. Heat shock transcription factor 1 (HSF1) known exert neuroprotective effects on neurodegenerative diseases. This study sought analyse the role and mechanism of HSF1 in PD-induced The PD mouse models were established using 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP). Animal behaviour capacities neuronal damage assessed via behavioural tests, tyrosine hydroxylase (TH) staining,...
Automated polyp segmentation technology plays an important role in diagnosing intestinal diseases, such as tumors and precancerous lesions.Previous works have typically trained convolution-based U-Net or Transformer-based neural network architectures with labeled data.However, the available public datasets are too small to train sufficiently, suppressing each network's potential performance.To alleviate this issue, we propose a universal data augmentation synthesize more from existing...
Action recognition in still images has been recently promoted by deep learning. However, the success of these models heavily depends on huge amount training for various action categories, which may not be available practice. Alternatively, humans can classify new categories after seeing few images, since we only compare appearance similarities between hand, but also attempt to recall importance motion cues from relevant videos our memory. To mimic this capacity, propose a novel Hybrid Video...
Video captioning is a popular task which automatically generates natural-language sentence to describe video content. Previous works mainly use the encoder–decoder framework and exploit special techniques such as attention mechanisms improve quality of generated sentences. In addition, most focus on global features spatial features. However, are usually fully connected Recurrent convolution networks (RCNs) receive 3-dimensional input at each time step, but temporal structure channel step has...
In supervised learning, a well-trained model should be able to recover ground truth accurately, i.e. the predicted labels are expected resemble as much possible. Inspired by this, we formulate difficulty criterion based on recovery degrees of training examples. Motivated intuition that after skimming through corpus, neural machine translation (NMT) “knows” how schedule suitable curriculum according learning difficulty, propose self-guided strategy encourages NMT learn from easy hard basis...