A5100446703- Speech and Audio Processing
- EEG and Brain-Computer Interfaces
- Music and Audio Processing
- Blind Source Separation Techniques
- Speech Recognition and Synthesis
- Emotion and Mood Recognition
- Medical Image Segmentation Techniques
- Neural dynamics and brain function
- Text and Document Classification Technologies
- Spectroscopy and Chemometric Analyses
- Advanced Neural Network Applications
- Hearing Loss and Rehabilitation
- Video Surveillance and Tracking Methods
- Music Technology and Sound Studies
- Brain Tumor Detection and Classification
- Bayesian Methods and Mixture Models
- Advanced Image and Video Retrieval Techniques
- Drug Solubulity and Delivery Systems
- Advanced Drug Delivery Systems
- Advanced Image Processing Techniques
- Gaze Tracking and Assistive Technology
- Optical Coherence Tomography Applications
- Industrial Vision Systems and Defect Detection
- Advanced Computing and Algorithms
- Neuroscience and Music Perception
East China University of Science and Technology
2014-2025
Xi'an Polytechnic University
2024
Jimei University
2009-2012
Institute of Mechanics
2006
New Mexico State University
2001
Abstract Objective . To take full advantage of both labeled data and unlabeled ones, the Graph Convolutional Network (GCN) was introduced in electroencephalography (EEG) based emotion recognition to achieve feature propagation. However, a single cannot represent emotional state entirely precisely due instability EEG signal complexity state. In addition, noise existing graph may affect performance greatly. solve these problems, it necessary introduce feature/similarity fusion reduction...
In multi-label learning, leveraging contrastive learning to learn better representations faces a key challenge: selecting positive and negative samples effectively utilizing label information. Previous studies selected based on the overlap between labels used them for label-wise loss balancing. However, these methods suffer from complex selection process fail account varying importance of different labels. To address problems, we propose novel method that improves through distribution....
Abstract Objective . The instability of the EEG acquisition devices may lead to information loss in channels or frequency bands collected EEG. This phenomenon be ignored available models, which leads overfitting and low generalization model. Approach Multiple self-supervised learning tasks are introduced proposed model enhance emotion recognition reduce problem some extent. Firstly, channel masking simulate certain resulting from EEG, two learning-based feature reconstruction combining...
Abstract Objective . Auditory attention decoding (AAD) determines which speaker the listener is focusing on by analyzing his/her EEG. Convolutional neural network (CNN) was adopted to extract spectro-spatial-feature (SSF) from short-time-interval of EEG detect auditory spatial without stimuli. However, following factors are not considered in SSF-CNN scheme. (a) Single-band frequency analysis cannot represent pattern precisely. (b) The power feature related dynamic patterns attended stimulus....
In this letter, we present a new speech hash function based on the non-negative matrix factorization (NMF) of linear prediction coefficients (LPCs). First, analysis is applied to obtain its LPCs, which represent frequency shaping attributes vocal tract. Then, NMF performed LPCs capture speech's local feature, then used for vector generation. Experimental results demonstrate effectiveness proposed in terms discrimination and robustness against various types content preserving signal...
Various musical descriptors have been developed for Cover Song Identification (CSI). However, different are based on various assumptions, designed representing distinct characteristics of music, and often differ in scale noise level. Therefore, a single similarity function combined with specific descriptor is generally not able to describe the between songs comprehensively reliably. In this paper, we propose two-layer fusion model CSI, which combines information carried by functions...
ABSTRACT Accurate segmentation of the optic cup and disc in fundus images is crucial for prevention diagnosis glaucoma. However, challenges arise due to factors such as blood vessels, mainstream networks often demonstrate limited capacity extracting contour information. In this paper, we propose a framework named FDT‐Net, which based on frequency‐aware dual‐branch Transformer (FDBT) architecture with parallel information mining uncertainty‐guided refinement. Specifically, design FDBT that...
Similarity measurement plays an important role in various information retrieval tasks. In this paper, a music scheme based on two-level similarity fusion and post-processing is proposed. At the level, to take full advantage of common complementary properties among different descriptors functions, first, track-by-track graphs generated from same descriptor but functions are fused with network (SNF) technique. Then, obtained first-level similarities further mixture Markov model (MMM) diffusion...