Jun Luo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jun Luo has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Mingliang Zhou
School of Computer Science, Chongqing University, Chongqing, China
3 shared publications - Xuekai Wei
School of Computer Science, Chongqing University, Chongqing 400044, China
3 shared publications - Huayan Pu
Chongqing University
2 shared publications - Guangchao Yang
Chongqing University
1 shared publication - Ji Chen
Purple Mountain Laboratories
1 shared publication - Jinsong Wu
College of Computer Science, Chongqing University, Chonqqing 400044, P. R. China
1 shared publication
Latest Publications
- 21 July 2025
- Published by World Scientific Pub Co Pte Ltd in Journal of Circuits, Systems and Computers
- Vol. 34 (18)
Real-Time Time Series-Based Behavior Anomaly Detection and Strategic Innovation in Edge Teaching Environments
With the development of educational informatization, multimedia teaching has become an essential means of higher education and has penetrated various disciplines. Traditional teaching methods have difficulty detecting and effectively intervening in students’ abnormal behavior in time, which affects students’ academic performance and future development. Recently, deep learning algorithms have been employed for abnormal behavior detection. However, existing studies focus on prediction accuracy and lack solutions and dynamic analysis. To further promote the development of students’ abnormal behavior detection technology and improve teaching quality, this paper proposes a time series-based framework for students’ abnormal behavior detection and learning behavior recommendation, which can dynamically identify and monitor students’ abnormal behavior. Specifically, LSTM and the transformer are used in this paper to capture the time dependence of students’ learning behaviors for binary classification prediction. Variational autoencoders (VAEs) are used as generative models to learn the distribution of learning behaviors of low-risk students and recommend learning behaviors for high-risk students. The experimental results show that the framework proposed in this paper can realize real-time monitoring and personalized intervention of students’ learning behavior, which is helpful for reforming teaching and improving teaching quality.
- 21 July 2025
- Published by World Scientific Pub Co Pte Ltd in Journal of Circuits, Systems and Computers
- Vol. 34 (18)
Real-Time Time Series-Based Behavior Anomaly Detection and Strategic Innovation in Edge Teaching Environments
With the development of educational informatization, multimedia teaching has become an essential means of higher education and has penetrated various disciplines. Traditional teaching methods have difficulty detecting and effectively intervening in students’ abnormal behavior in time, which affects students’ academic performance and future development. Recently, deep learning algorithms have been employed for abnormal behavior detection. However, existing studies focus on prediction accuracy and lack solutions and dynamic analysis. To further promote the development of students’ abnormal behavior detection technology and improve teaching quality, this paper proposes a time series-based framework for students’ abnormal behavior detection and learning behavior recommendation, which can dynamically identify and monitor students’ abnormal behavior. Specifically, LSTM and the transformer are used in this paper to capture the time dependence of students’ learning behaviors for binary classification prediction. Variational autoencoders (VAEs) are used as generative models to learn the distribution of learning behaviors of low-risk students and recommend learning behaviors for high-risk students. The experimental results show that the framework proposed in this paper can realize real-time monitoring and personalized intervention of students’ learning behavior, which is helpful for reforming teaching and improving teaching quality.
- 10 April 2025
- Published by World Scientific Pub Co Pte Ltd in Journal of Circuits, Systems and Computers
- Vol. 34 (09)
Parallax Compensation-Driven Feature Transform Network for Super-Resolution Image Quality Assessment
With the development of super-resolution (SR), how to assess the quality of SR-reconstructed images fairly and reliably is a realistic and critical challenge. However, most existing easily portable blind SR image quality assessment (SR-IQA) methods consider only multi-scale depth information in a single domain, neglecting the utilization of multi-scale information in different domains. In addition, the technique of merely altering the feature extraction block does not accurately replicate how the human visual system (HVS) perceives quality. To address these problems, in this paper, we introduce a parallax compensation-driven feature transform for the super-resolution image quality assessment (PCFTSR-IQA) metric. First, the multi-level feature extraction framework based on the Karhunen–Loève transform (KLT) is specifically presented to transform the inputs to the feature domain for better extraction of structural and detailed features. Second, a set of parallax compensation loss functions is designed to replicate the HVS perception mechanism, which employs parallax compensation to perform constraints and filtering during the feature extraction process for maximum feature utilization. Finally, we train the end-to-end PCFTSR-IQA network for no-reference (NR) SR-IQA. Extensive experiments suggest that the metric we propose exceeds the performance of leading SR-IQA metrics on public datasets.
- 10 April 2025
- Published by World Scientific Pub Co Pte Ltd in Journal of Circuits, Systems and Computers
- Vol. 34 (09)
Parallax Compensation-Driven Feature Transform Network for Super-Resolution Image Quality Assessment
With the development of super-resolution (SR), how to assess the quality of SR-reconstructed images fairly and reliably is a realistic and critical challenge. However, most existing easily portable blind SR image quality assessment (SR-IQA) methods consider only multi-scale depth information in a single domain, neglecting the utilization of multi-scale information in different domains. In addition, the technique of merely altering the feature extraction block does not accurately replicate how the human visual system (HVS) perceives quality. To address these problems, in this paper, we introduce a parallax compensation-driven feature transform for the super-resolution image quality assessment (PCFTSR-IQA) metric. First, the multi-level feature extraction framework based on the Karhunen–Loève transform (KLT) is specifically presented to transform the inputs to the feature domain for better extraction of structural and detailed features. Second, a set of parallax compensation loss functions is designed to replicate the HVS perception mechanism, which employs parallax compensation to perform constraints and filtering during the feature extraction process for maximum feature utilization. Finally, we train the end-to-end PCFTSR-IQA network for no-reference (NR) SR-IQA. Extensive experiments suggest that the metric we propose exceeds the performance of leading SR-IQA metrics on public datasets.
- 3 April 2024
- Published by World Scientific Pub Co Pte Ltd in Journal of Circuits, Systems and Computers
- Vol. 33 (14)
Deep Fusion Module for Video Action Recognition
In video action recognition, effective spatiotemporal modeling is crucial. However, traditional two-stream methods face challenges in integrating spatial information from RGB images and temporary information from optical flow without long-range temporal modelling. To address these limitations, we propose the Deep Fusion Module (DFM), which focuses on the deep fusion of spatial and temporal information and consists of two components. First, we propose an Attention Fusion Module (AFM) to effectively fuse the shallow features obtained from a two-stream network, thereby facilitating the integration of spatial and temporal information. Next, we incorporate a SpatioTemporal Module (STM), comprising a ConvGRU and a 1×1 convolution, to model long-range temporal dependency and fuse spatial-temporal features. Experiments on the UCF101 dataset show that our method achieves 96.5% accuracy, outperforming baseline two-stream models by 0.3%.
- 3 April 2024
- Published by World Scientific Pub Co Pte Ltd in Journal of Circuits, Systems and Computers
- Vol. 33 (14)
Deep Fusion Module for Video Action Recognition
In video action recognition, effective spatiotemporal modeling is crucial. However, traditional two-stream methods face challenges in integrating spatial information from RGB images and temporary information from optical flow without long-range temporal modelling. To address these limitations, we propose the Deep Fusion Module (DFM), which focuses on the deep fusion of spatial and temporal information and consists of two components. First, we propose an Attention Fusion Module (AFM) to effectively fuse the shallow features obtained from a two-stream network, thereby facilitating the integration of spatial and temporal information. Next, we incorporate a SpatioTemporal Module (STM), comprising a ConvGRU and a 1×1 convolution, to model long-range temporal dependency and fuse spatial-temporal features. Experiments on the UCF101 dataset show that our method achieves 96.5% accuracy, outperforming baseline two-stream models by 0.3%.