Author: School of Artificial Intelligence and Automation Edited by: Chen Machuan

WUST News - Professor Li Weigang's team at the School of Artificial Intelligence and Automation, Wuhan University of Science and Technology (WUST) has achieved remarkable success in the field of intelligent sensing. Doctoral students Tian Zhiqiang, Xie Lu, Wang Yongqiang, and Wang Qifeng, each serving as the first author with Professor Li Weigang as the corresponding author (second author), have consecutively published five high-level papers in prestigious international journals, including Pattern Recognition (CAS Tier 1 TOP), Applied Soft Computing (CAS Tier 2 TOP), IEEE Transactions on Vehicular Technology (CAS Tier 2), and IEEE Transactions on Instrumentation and Measurement (CAS Tier 2). These publications highlight the team's ongoing innovation and research influence in intelligent sensing, 3D information processing, and robotic perception systems.

In Pattern Recognition, Tian Zhiqiang addressed the vulnerability of 3D point cloud data in complex environments by proposing a novel "desensitized adversarial training" method. This approach optimizes neural networks from the perspective of feature sensitivity, significantly enhancing the model's robustness to damaged point clouds and achieving excellent results on public datasets.

In Applied Soft Computing, Xie Lu introduced a new semantic segmentation method called HEL-NC, which innovatively combines neural collapse theory with hard sample learning. By employing an equiangular tight frame classifier and a weighted loss function, the method effectively improves segmentation accuracy on imbalanced class data, demonstrating broad engineering application potential.

Wang Yongqiang proposed a robust skeleton registration framework (SRRF) in Applied Soft Computing, which innovatively incorporats interference-resistant skeleton structures into point cloud registration. By combining a distribution distance loss function to strengthen skeleton consistency, the framework effectively mitigates registration errors caused by noise, uneven density, and geometric deformation, showcasing strong robustness and practical application value.

In IEEE Transactions on Vehicular Technology, Wang Qifeng introduced the MTC-SLAM system, which incorporates a multi-scale tightly coupled SLAM and loop closure detection method. This significantly enhances the positioning accuracy and robustness of SLAM systems in dynamic and complex environments. Additionally, he published new findings in IEEE Transactions on Instrumentation and Measurement, presenting an ICP registration strategy based on reliable initial values and adaptive thresholds, substantially improving the positioning accuracy of laser odometry in complex dynamic environments.

Professor Li Weigang's team has long focused on interdisciplinary areas such as industrial process control, intelligent detection, artificial intelligence and machine learning algorithms, mobile robotics, and intelligent driving. The team is dedicated to cultivating high-level young research talent and continuously contributing to innovation in industrial intelligent technology. (School of Artificial Intelligence and Automation)