海外英才分论坛学术报告【Deep generative adversarial learning for crack-like object detection】
时间：2019年5月14日（周二）上午9:30
地点：仓山校区光电学院三楼304会议室
主讲：犹他大学计算机系张凯歌博士
主办：福建师范大学光电与信息工程学院、医学光电科学与技术教育部重点实验室、福建省光子技术重点实验室、福建省光电传感应用工程技术研究中心
报告摘要:
During the past few years, deep learning has achieved great success and has been utilized for solving a variety of object detection problems successfully. However, using deep learning for high accurate crack localization is non-trivial. First, region based object detection cannot locate cracks accurately, and it is very inefficient. Second, the object detection networks have severe data imbalance problem inherent in crack-like object detection which can fail the training. Third, deep learning based methods are domain sensitive, which results in poor model-generalization ability. Fourth, deep learning is a data driven method which relies on a large amount of manually labeled ground truths (GTs) for the training that is labor-intensive and even infeasible, especially for annotating pixel-level GTs with low-resolution crack images. This research has proposed an efficient deep neural network based on generative adversarial learning which solves all the abovementioned problems..
专家简介:
Kaige Zhang is a research assistant at CVPRIP (computer vision, pattern recognition and image processing) lab, Utah State University. He received his B.S. in electronic engineering from Harbin Institute of Technology, Harbin, China, in 2011, and M.S. degree in signal and information processing from Harbin Engineering University, Harbin, China, in 2014. His research interests include computer vision, machine learning and data mining. During his Ph.D. study, he has focused on designing efficient deep neural networks for crack-like object detection which has a wide range of industrial applications such as pavement surface inspection, bridge cracking monitor, railway track assessment, etc. He has proposed a suite of algorithms for solving many key problems in crack-like object detectoin, such as data imbalance, noise issue, algorithm robustness, efficiency of deep network, etc. From May to September 2018, he was a research engineer at Hivision research (Silicon Valley) where he helped the company solved a seriers of problems in robust face recognition. He has been a session chair for 2018 IEEE international conference on intelligent transportation systems (ITSC). He has been a reviewer for International Journal of Pavement Research and Technology (Elsevier), IEEE Access, ITSC2018, Recent Patents on Computer Science, etc.
