A5101622539- Human Pose and Action Recognition
- Gait Recognition and Analysis
- Anomaly Detection Techniques and Applications
- Web Data Mining and Analysis
- Text and Document Classification Technologies
- Service-Oriented Architecture and Web Services
- Domain Adaptation and Few-Shot Learning
- Color Science and Applications
- Image Enhancement Techniques
- Image and Signal Denoising Methods
- Face and Expression Recognition
- Caching and Content Delivery
- Multimodal Machine Learning Applications
- Hand Gesture Recognition Systems
- Video Surveillance and Tracking Methods
- Spectroscopy and Chemometric Analyses
- Mobile Crowdsensing and Crowdsourcing
- Web Applications and Data Management
- Neural dynamics and brain function
- Scheduling and Timetabling Solutions
- Advanced Computing and Algorithms
- Blind Source Separation Techniques
- Constraint Satisfaction and Optimization
- Smart Agriculture and AI
- Security in Wireless Sensor Networks
Hunan University of Science and Technology
2023-2024
National Central University
2024
Guangzhou University
2021-2024
Zhejiang University of Technology
2023
Xi'an University of Technology
2006-2021
Institute of Software
2021
Jiangxi Applied Engineering Vocational Institute
2020
South China Agricultural University
2018
Hong Kong University of Science and Technology
2016
University of Hong Kong
2016
In this paper, we propose a three-stream convolutional neural network (3SCNN) for action recognition from skeleton sequences, which aims to thoroughly and fully exploit the data by extracting, learning, fusing inferring multiple motion-related features, including 3D joint positions displacements across adjacent frames as well oriented bone segments. The proposed 3SCNN involves three sequential stages. first stage enriches independently extracted features co-occurrence feature learning....
The application of optical remote sensing images (ORSIs) is prevalent in many fields. Accordingly, ORSI-oriented salient object detection (SOD) has attracted more attention recent years. However, yet previously proposed methods present appealing performance natural scene (NSIs), they are difficult to be directly extended due the complex scenes, such as blended backgrounds and diversiform topological shapes. Most specifically designed models often fail achieve satisfactory results weak usage...
Various techniques have been proposed to detect smells in spreadsheets, which are susceptible errors. These typically spreadsheet through a mechanism based on fixed set of patterns or metric thresholds. Unlike conventional programs, tabulation styles vary greatly across spreadsheets. Smell detection thresholds, insensitive the varying styles, can miss many one while reporting spurious another. In this paper, we propose CUSTODES effectively cluster cells and these clusters. The clustering...
Reading requires appropriate strategies to spark initial interest and sustain engagement. One promising strategy is the pedagogical approach of learning-by-teaching, transforming learners into active participants. Integrating this digitalized individualized reading contexts has potential foster development young readers. Currently, AI techniques are primarily used in chatbots as tutors, with limited focus on tutee that employ learning-by-teaching pedagogy. Therefore, study adopted a...
Abstract The purpose of 3D human pose estimation is to estimate the coordinates key points body directly from images. Although multi‐view based methods have better performance and higher precision coordinate than a single‐view based, they need know camera parameters. In order effectively avoid restriction this constraint improve generalizability model, dual‐view single‐person method without parameters proposed. This first uses 2D network HR‐net joint point two images with different views,...
With the wide adoption of Web APIs released on Internet, users tend to reuse them for business requirements or software development. Mashup is a useful technology composing into new and value-added application. increasing number Mashups, API-Mashup ecosystem has emerged based invocation relationship between Mashups APIs. In this paper, we take ProgrammableWeb, typical ecosystem, as an example investigate its dynamic evolutionary analysis. Although there have been some works they mainly focus...
Suffering from the multi-view data diversity and complexity for semi-supervised classification, most of existing graph convolutional networks focus on architecture construction or salient structure preservation, ignore complete classification contribution. To mine more distribution with consideration specificity commonality, we propose fusion based (SF-GCN) improving performance classification. SF-GCN can not only retain special characteristic each view by spectral embedding, but also...
Apoptosis and fatty acid-binding protein-4 (FABP4) induced-endoplasmic reticulum (ER) stress in macrophage is an important pathological process several vascular occlusive diseases, including atherosclerosis, both of which are accelerated by lipids or inflammatory cytokines.To determine whether interleukin 17A (IL-17A) accelerates atherosclerosis through activating FABP4-mediated ER macrophage.We show here that IL-17A induced murine human-derived macrophages vitro, the atherosclerotic lesions...
The Non-Local Means (NLM) filter uses the redundancy of information in image to remove noise, this scheme gives some best results among other powerful methods such as wavelet based approaches or diffusion techniques. Though simple implement and efficient practice, classical NLM suffers from ringing artifacts around edges when using square patches, due an abrupt lack image. This paper presents extended on Multi-Shape Patches Aggregation (NLM-MSPA) overcome problem, it medical ultrasound...
This paper presents a new approach to face recognition using independent component analysis (ICA) and fuzzy support vector machine (FSVM). Firstly, 2D wavelet transform is adopted obtain different level of coefficients. Secondly ICA applied on the low-frequency which contains most discrimination information original image. One criterion that not all ICs are useful for demonstrated rule selecting proposed. For reducing computational cost, fast method Then, FSVM classifier designed...
Feature generating networks face to the most important question, which is fitting difference (inconsistence) of distribution between generated feature and real data. This inconsistence further influence performance model, because training samples from seen classes disjointed with testing unseen in zero-shot learning (ZSL). In generalization (GZSL), come not only but also for closer practical situation. Therefore, difficultly obtain satisfactory challenging GZSL by adversarial semantic...
With the development of Mashup technique, number Web APIs released on continues to grow year by year. However, it is a challenging issue find and select desirable among large amount APIs. Consequently, interactive API recommendation used alleviate difficulty service selection, when users or developers try invoke for solving their business requirements software requirements. Currently, there are several collaborative filtering based approaches proposed recommendation, while performance...
Existing zero-shot learning (ZSL) methods usually learn a projection function between feature space and semantic embedding space(text or attribute space) in the training seen classes testing unseen classes. However, cannot be used multi-semantic spaces, which have diversity characteristic for describing different information of same class. To deal with this issue, we present novel method to ZSL based on class label autoencoder (CLA). CLA can not only build uniform framework adapting but also...
Abstract Cross‐domain action recognition is a challenging vision task due to the domain shift and absence of labeled data in target domain. With only labelled source unlabelled during training, some existing methods rely on an adversarial framework align features from different domains common latent space. However, adversarial‐based approaches have major limitation attempting perform alignment holistic view, ignoring underlying coherence class structure across domains. A structure‐aware loss...
We present an interface-type memristor utilizing native silicon oxide nanostructures achieved through controlled plasma oxidation engineering at room temperature. This process effectively reduces the concentration of oxygen vacancies on surface oxide, facilitating oxygen-vacancy-mediated interfacial resistive switching with excellent COMS compatibility and low variability (1.03% for C2C 3.96% D2D variability). Moreover, this dynamic exhibits favorable nonlinearity short-term memory (STM)...
This work presents a novel approach to multimodal human action recognition by jointly using visual RGB and depth (including skeleton joint positions) data captured from camera. For the feature extraction, Local Surface Geometric Feature (LSGF) is adopted capture geometric appearance postures cues. Simultaneously, improved dense trajectory (IDT) extracted modality characterize motion, shape information. These features different modalities are complementary each other. Then two-stage...