Li Zhao

Li Zhao

Auto-generated profile, based on publicly available information

Li Zhao

National Engineering Research Center for E-learning, Central China Normal University, China

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
26 April 2024
Verified
Conference Paper
Open access
HLHai Liu
Hai Liu
YSYu Song
Yu Song
Tingting LiuTingting Liu
Tingting Liu
HLHui Li
Hui Li
JLJiawen Li
Jiawen Li
LZLi Zhao
Li Zhao
In the infrared spectral domain, the presence of random noise and the overlapping of adjacent peaks pose significant challenges. To tackle these issues, we have introduced a novel model for the reconstruction of infrared spectra, incorporating enhancements to the total variation regularization. The maximum a posteriori (MAP) theory is introduced to construct the signal reconstruction model for latent infrared spectrum. The proposed model can remove the random spectral noises and produce the new peaks. In the MAP architecture, the prior probability is modeled as the improved total variation constraint, the Gaussian distribution-based spectral noise is constructed for the likelihood probability distribution. Then, the proposed infrared spectral signal model is optimized by the famous split Bregman iteration approach. The high-quality infrared spectrum and instrument function can be estimated simulated by the proposed model on the simulated and real spectral signals. All the comparative experiments were conducted using Matlab 2022, illustrating the effective performance in suppressing spectral noises and splitting overlap peaks.

26 April 2024
Verified
Conference Paper
Open access
HLHui Li
Hui Li
JLJiawen Li
Jiawen Li
HLHai Liu
Hai Liu
LZLi Zhao
Li Zhao
Tingting LiuTingting Liu
Tingting Liu
XYXinge You
Xinge You
Speech emotion recognition plays an important role in the field of human speech communication. Nonetheless, the ensuing speech frequently encounters challenges characterized by indistinct emotional expressions and sporadic interference from random noise. In this paper, to effectively tackle these challenges, we propose a novel few critical cues Transformer, which learns a few core features and long-distance features in speech. To achieve this objective, two novel modules leveraging speech emotion features are introduced: the stage feature module and the critical cues module. Using the Transformer as the backbone, long-term dependencies in the speech graph are revealed. The proposed method is carried out on several speech emotion datasets, such as IEMOCAP and EMO-DB. Experimental results substantiate that the proposed FC2VR2 method can achieve better performance than other state-of-the-art methods.

26 April 2024
Verified
Conference Paper
Open access
HLHua Liu
Hua Liu
LZLi Zhao
Li Zhao
XHXin He
Xin He
HLHai Liu
Hai Liu
Feedback from teaching agents is beneficial for enhancing learning performance, and the emotional design of feedback can effectively evoke positive emotions in learners and enhance their learning motivation. However, there is still limited research on feedback strategies of teaching agents supported by emotional design, addressing issues such as low learner engagement and inhibition of collaborative knowledge construction in online collaborative sessions. By reviewing relevant studies on emotional response theory, this article constructs a model of feedback from teaching agents supported by emotional design, providing a reference for the design of intelligent teaching agent feedback for online collaborative sessions. Based on this, the article conducts a quasi-experimental study to validate the effectiveness of the model. The study finds that: (1) Feedback on learning situations provided by teaching agents can promote participation in collaborative sessions and collaborative knowledge construction; (2) Feedback from teaching agents supported by emotional design is more effective in promoting participation in online collaborative sessions and collaborative knowledge construction; (3) Learners’ technical acceptance of teaching agents providing feedback on learning situations and positive emotions is significantly higher than that of teaching agents providing only feedback on learning situations. These findings provide valuable guidance for the design and development of teaching agents.

26 April 2024
Verified
Conference Paper
Open access
LZLi Zhao
Li Zhao
SLShuangshuang Li
Shuangshuang Li
XHXin He
Xin He
HLHai Liu
Hai Liu
While current deep learning-based methods for image emotion classification have achieved superior performance, most approaches primarily extract generic features directly from the entire emotional image, without considering the unique process of emotional arousal in human cognition. Addressing this issue and inspired by the "stimulus-organism-response" psychological model, this paper introduces a multi-incentive perception network to simulate the emotional arousal process, aiming to tackle the challenge of abstraction in image emotion analysis.Drawing from relevant psychological studies, the paper selects color, objects, and faces as emotional stimuli and designs three specific networks to selectively extract emotional features from different stimuli. Additionally, the paper proposes a hierarchical cross-entropy loss to distinguish between simple negative samples and challenging negative samples. The performance of this method consistently surpasses existing approaches on commonly used datasets for image emotion classification, and visual results also demonstrate the importance of various incentives in the emotional arousal process.

26 April 2024
Verified
Conference Paper
Open access
Tingting LiuTingting Liu
Tingting Liu
HLHai Liu
Hai Liu
JLJiawen Li
Jiawen Li
LZLi Zhao
Li Zhao
HLHui Li
Hui Li
TFTao Feng
Tao Feng
Infrared facial expression classification often suffers from the issues of low-quality facial images, subjectivity of annotators and human expressions ambiguity. To tackle the aforementioned challenges, we introduce a novel approach for facial expression classification leveraging Gaussian label distribution learning and graph neural networks. This method aims to enhance the robustness and accuracy of expression recognition by effectively modeling the uncertainty inherent in facial expressions and capturing complex dependencies among facial features. Specially, the graph convolutional networks are introduced to achieve relationship between action units (AUs) and expressions by learning AUs graph representation. The hard expression labels are constructed by the cues, which are produced by the graph convolutional networks. The constructed sentiment label distribution is subsequently trained using a CNN framework for facial expression classification in an end-to-end manner. Experimental findings across multiple publicly available datasets substantiate the efficacy of the proposed methodology, showcasing its superior performance when juxtaposed against other contemporary approaches in the field.