Invited Speaker丨特邀报告

 

 Assoc. Prof. Tingru ZHANG, Shenzhen University, China

Tingru Zhang is an associate professor at the Institute of Human Factors and Ergonomics, Shenzhen University, and Guest Researcher at the CGN National Key Laboratory. Her main research focuses on intelligent transportation, human-AI interaction, and smart interaction & user experience. She has led over 10 research projects. With more than 50 publications, her work includes 1 ESI Hot Paper and 1 Highly Cited Paper. She currently serves as an editorial board member for Transportation Research Part F and International Journal of Ergonomics, and has been recognized with honors such as the Top 2% Most-Cited Scientists Worldwide (2023, 2024) and the Outstanding Early Career Award from the International Ergonomics Association (2023).

 

Speech Topic: EEG-based assessment of driver trust

Abstract: Effective collaboration between automated vehicles (AVs) and human drivers relies on maintaining an appropriate level of trust. However, real-time assessment of human trust remains a significant challenge. While initial efforts have delved into the potential use of physiological signals, such as skin conductance and heart rate, to evaluate trust, limited attention has been given to the feasibility of assessing trust through electroencephalogram (EEG) signals. This study aimed to address this issue by using EEG signals to objectively assess driver trust towards AVs. A total of 420 time- and frequency-domain EEG features were extracted, and nine machine learning algorithms were applied to construct driver trust assessment models. Additionally, to explore the potential of developing cost-effective models with reduced feature inputs, this study developed trust models using features solely from single brain regions: frontal, parietal, occipital, or temporal. The results showed that the best-performing model, utilizing features from the whole brain and employing the Light Gradient Boosting Machine (LightGBM) algorithm, achieved an accuracy of 88.44% and an F1-score of 78.31%. In comparison, models based on single brain regions did not achieve comparable performance to the comprehensive model. However, the frontal and parietal regions showed important potentials for developing costeffective trust assessment models. This study also performed feature analysis on the best-performing model to identify features highly responsive to changes in trust. The results showed that an increased power of beta waves tended to indicate a lower level of trust in AVs. These findings contribute to our understanding of the neural correlates of trust in AVs and hold practical implications for the development of trust-aware AV technologies capable of adapting and responding to driver’s trust levels effectively.

Asst. Prof. Wenyu JIANG, Shenzhen University, China

Dr. Jiang Wenyu is an Assistant Professor and Master's Supervisor at Shenzhen University, holding a Ph.D. in Engineering Physics from Tsinghua University. His research focuses on intelligent monitoring of transportation infrastructure, computer vision & spatial intelligence, and disaster reduction & emergency management. He has led/participated in multiple national and provincial research projects, including the National Key R&D Program of China, and Guangdong Provincial Key R&D Initiatives. As the first/corresponding author, he has published 20+ papers in top-tier journals such as International Journal of Applied Earth Observation and Geoinformation, Environmental Modelling & Software, and International Journal of Disaster Risk Reduction. He holds 5 invention patents and 5 software copyrights, and was twice awarded the Grand Prize of Guangdong Provincial Science & Technology Progress Award in Emergency & Safety (2023, 2024). His innovations have been deployed by emergency management agencies in Guangdong, Sichuan, Zhejiang, and Heilongjiang provinces, providing critical technical support for disaster management.

 

Speech Topic: UAV-Based Intelligent Vision Framework for Disaster Risk Early-Warning: A Case Study of Wildfire Identification

Abstract: Low-altitude UAVs serve as critical aerial transportation platforms with significant potential in emergency management due to superior maneuverability, flexibility, and visual coverage. However, motion blur induced by high-speed flight severely degrades visual image quality, constraining practical deployment efficacy. This study develops an UAV-based intelligent visual warning framework, with wildfire identification as a representative case. The FLAME benchmark dataset for drone-based wildfire detection is employed, and a physics-driven blurring algorithm simulates image degradation under varying flight parameters. With ResNet-50 and Vision Transformer (ViT) as baseline models, transfer learning is introduced to enhance motion-blur robustness. Results reveal that on pristine data, ViT (OA: 99.67%, F1: 99.73%) slightly outperforms ResNet-50 (OA: 98.50%, F1: 98.81%). Under motion degradation (motion index=25), ViT (OA: 88.56%) shows significant superiority over ResNet-50 (OA: 82.93%, ~6%↓). Furthermore, the transfer-enhanced ViT-Pre model (OA: 92.47%) achieves a 3.91-point OA gain versus native ViT (~4%↓), demonstrating the efficacy of transfer learning in compensating motion-blur-induced accuracy loss. This work provides practical insights for robust vision system design in dynamic UAV-based disaster risk early-warning systems.