A5100411934- Face and Expression Recognition
- Remote-Sensing Image Classification
- Advanced Image and Video Retrieval Techniques
- Advanced Computing and Algorithms
- Video Surveillance and Tracking Methods
- Sparse and Compressive Sensing Techniques
- Image Processing and 3D Reconstruction
- Blind Source Separation Techniques
- Image and Object Detection Techniques
- Network Security and Intrusion Detection
- Text and Document Classification Technologies
- Spam and Phishing Detection
- Medical Image Segmentation Techniques
- Infrared Target Detection Methodologies
- Speech and Audio Processing
- Image Processing Techniques and Applications
- Advanced Algorithms and Applications
- Music and Audio Processing
- Tensor decomposition and applications
- Video Analysis and Summarization
- Advanced Data Compression Techniques
- Neural Networks and Applications
- Speech Recognition and Synthesis
- Moringa oleifera research and applications
- Handwritten Text Recognition Techniques
State Administration of Cultural Heritage
2023-2025
Northwest University
2025
Agency for Science, Technology and Research
2016-2024
Tianjin University of Finance and Economics
2023-2024
Sichuan University
2023
Yunnan University
2016-2023
Science and Technology on Surface Physics and Chemistry Laboratory
2023
China Electronics Technology Group Corporation
2023
Cultural Relics Institute Hebei Province
2018-2022
The University of Texas Health Science Center at San Antonio
2022
We introduce a new benchmark "Humans Interacting with Common Objects" (HICO) for recognizing human-object interactions (HOI). demonstrate the key features of HICO: diverse set common object categories, list well-defined, sense-based HOI and an exhaustive labeling co-occurring category in each image. perform in-depth analysis representative current approaches show that DNNs enjoy significant edge. In addition, we semantic knowledge can significantly improve recognition, especially uncommon categories.
In real-world scenarios, the number of phishing and benign emails is usually imbalanced, leading to traditional machine learning or deep algorithms being biased towards misclassifying emails. Few studies take measures address imbalance between them, which significantly threatens people’s financial information security. To mitigate impact on model enhance detection performance emails, this paper proposes two new with undersampling: Fisher–Markov-based ensemble (FMPED) method...
Knowledge-based questions are typically employed to evaluate LLM's knowledge boundaries; meanwhile, numerous studies focus on question generation as a means enhance the capabilities of both models and individuals. However, there is lack in-depth exploration about what constitutes good from perspective cognition. This paper proposes aligning complete underlying with educational criteria effectively in physics courses, thereby developing novel knowledge-intensive metrics quality. To this end,...
ABSTRACT Despite the large amount of video data captured during ethological studies wild mammals, there is no widely accepted method available to automatically and quantitatively measure analyze animal behavior. We developed a framework using facial recognition deep learning track, measure, quantify behavior single or multiple individuals from 10 distinct mammalian taxa, including three species primates, bovids, carnivores, one equid. used spatiotemporal information based on skeleton models...
The stitching of bone stick fragments is great significance for the inheritance and research into outstanding traditional culture Western Han Dynasty. Focused on problem that existing methods have a low fragment success rate due to their complex edges, this paper proposes method based improved corner detection, which strengthens features broken edges improves stitching. First, maximum outer contour obtained using connectivity analysis. Then, edge are enhanced by combining multi-scale...
Metal nitrides (FeCoN) surrounded by pyridinic N were identified as the active center for oxygen reduction reaction.
The rejoining of bone sticks holds significant importance in studying the historical and cultural aspects Han Dynasty. Currently, work inscriptions heavily relies on manual efforts by experts, demanding a considerable amount time energy. This paper introduces multi-scale feature fusion Siamese network guided edge contour (MFS-GC) model. Constructing framework, it first uses residual to extract features sticks, which is followed computing L2 distance for similarity measurement. During...
Tone mapping methods aim to compress the high dynamic range (HDR) images so that they can be displayed on common devices. The tone curve plays a key role in many methods, which directly adjust of HDR image. S-shaped curves produce impressive performances due their flexibility. However, conventional is single and had problem excessive compressing dense grayscale areas, resulting loss details this area, insufficient sparse low contrast mapped This paper proposes multi-peak (MPS) address these...
As the emergence of numerous services with similar functions, it is very helpful to recommend personalized for users, and urgent accurately predict QoS(Quality-of-Service) values Web services. Collaborative Filtering (CF) a commonly-used method handle above issues. However, faces two common issues: data sparsity problem trustworthiness issue, which greatly reduces its prediction accuracy. To address this properly systematically, we introduce network embedding learning into QoS process...
Abstract Comparative molecular field analysis and comparative similarity indices were employed to analyze the antiwear properties of a series 57 esters as potential lubricant-based oils. Predictive 3D-quantitative structure tribo-ability relationship models established using SYBYL multifit alignment rule with training set test set. The optimum all shown be statistically significant cross-validated coefficients q 2 > 0.5 conventional r 0.9, indicating that are sufficiently reliable for...
This paper presents a novel method for classifying regions from human movements in service robots' working environments. The entire space is segmented subject to the class type according functionality or affordance of each place which accommodates typical behavior. achieved based on grid map two steps. First probabilistic model developed capture cell by using non-ergodic HMM. Then learned transition probabilities corresponding these are used cluster all cells K-means algorithm. knowledge...
The diversity of multimedia data in the real world usually forms heterogeneous types feature sets. How to explore structure information and relationships among multiple features is still an open problem. In this paper, we propose unsupervised subspace learning method, named shared low-rank correlation embedding (SLRCE) for fusion. First, learned subspace, implement representation on each set enforce a constraint uncover common features. Second, develop enhanced analysis simultaneously...
The postMessage mechanism in HTML5 enables different webpage origins to exchange information and communicate. It becomes increasingly popular among the websites that need import contents from third-party services, such as advertisements preferable recommendations. Ideally, a receiver function should be locally implemented hosting page needs receive messages. However, real world, is usually provided by service provider, code imported via HTML "script" tag so deemed same origin with page. In...
In this article we consider Incremental Fisher linear discriminant (IFLD) based on data denoising. The denoising is completed by Markov sampling such that the generated non-noise sample sequence an uniformly ergodic chain (u.e.M.c.). We first establish generalization bounds of IFLD with u.e.M.c. samples, and prove algorithm samples consistent. also present two new classification algorithms sampling, (IFLD-MS) improved (IIFLD-MS). Experimental results benchmark repository suggest IFLD-MS...
Linear discriminant analysis (LDA) is an effective method for solving the classification problems. Many based-discriminant approaches have been proposed to extract more information and try overcome limitation of LDA. Local linear (LLDA) was capture local structure samples, it can assumption Gaussian distribution which emerge in traditional In this paper, we tensor version LLDA, tensorLLDA not only avoid undersampled problem appear LDA but also reduce computation complexity. Experiment on...
This paper presents a novel approach to modeling the dynamics of human movements with grid-based representation. The model we propose, termed as Multi-scale Conditional Transition Map (MCTMap), is an inhomogeneous HMM process that describes transitions location state in spatial and temporal space. Unlike existing work, our method able capture both local correlations long-term dependencies on faraway initiating events. enables learned incorporate more information generate informative...
Abstract Cross modal retrieval can retrieve images through a text query and vice versa. In recent years, cross has attracted extensive attention. The purpose of most now available methods is to find common subspace maximize the different correlation. To generate specific representations consistent with tasks, this paper proposes novel framework, which integrates feature learning latent space embedding. detail, we proposed deep CNN shallow extract samples. used representation images, uses...