讲座题目：Small Object Detection and Counting in Aerial Images Using Drones and Deep Learning 主讲人：Prof.Yi Shang 主持人：曹桂涛 教授 开始时间：2019-05-29 18:30:00 讲座地址：理科大楼B504 主办单位：计算机科学与软件工程学院
报告人简介：
Yi Shang is Professor and Director of Graduate Studies, Department of Electrical Engineering and Computer Science, University of Missouri, Columbia, Missouri. He received Ph.D. in Computer Science from University of Illinois at Urbana-Champaign in 1997, M.S. from the Institute of Computing Technology, Chinese Academy of Sciences, Beijing, in 1991, and B.S. from University of Science and Technology of China, Hefei, in 1988. He has published over 190 refereed papers in the areas of artificial intelligence, wireless sensor networks, mobile computing, and bioinformatics and has been granted 6 US patents. He has advised over 70 PhD and MS students. His research has been supported by NSF, NIH, US Department of Education, Army, DARPA, Microsoft, Raytheon, Missouri Department of Conservation, etc. Details of his lab, Distributed and Intelligent Computing Lab, can be found at http://dslsrv1.rnet.missouri.edu. 报告内容： Small object detection and counting in aerial images are active research and development areas and have many real applications. Monitoring waterfowl populations is essential for wildlife conservation in Missouri. This project aims at developing deep-learning based methods for small object (e.g., birds, people, animals, etc.) detection and counting in aerial images taken by drones or unmanned aircraft systems (UAS). The majority of the deep learning methods for object detection have been developed for large objects, and their performances on small-object detection are not as good. State-of-the-art deep learning methods for object detection are evaluated using a Little Birds in Aerial Imagery (LBAI) dataset, created from real-life aerial imagery data. They include object detection techniques YOLOv2, SSH, and Tiny Face, and small instance segmentation techniques U-Net and Mask R-CNN. SSH performed the best for easy cases, whereas Tiny Face performed the best for hard cases, where a cluttered background makes detecting birds difficult. |
