Satellite imaging technology has revolutionized geographic information systems. Jiang Li, Yan Lu, Krzysztof Koperski, and Chiman Kwan developed a fully convolutional neural network (FCN) model for refugee tent extraction. The Defense Advanced Research Project Agency supports the FCN project toward national security in gathering geographic spatial information to address effective refuge management in the southern border of Syria and Jordan.

The project "Deep Learning for Effective Refugee Tent Extraction Near Syria-Jordan Border " was published in the Institute of Electrical and Electronic Engineering (IEEE) Journal June 2020 issue. IEEE has more than 419,000 members in over 160 countries and is the most trusted "voice" for engineering, computing, and technology information.

According to the United Nations, the civil war in Syria forced more than 5.6 million people to flee their homes from the Islamic State of Iraq, and al-Sham (ISIS) held part of Syria to the neighboring countries seeking asylum. An influx of refugees between 40,000 and 50,000, mostly women and children, are desperately stranded in an isolated, formerly barren desert area known as Rukban.

According to the United Nations Institute for Training and Research Operational Satellite Applications Program (UNOSAT), the refugee population has increased exponentially within the last four years. There is an urgent need to manage the global refugee crisis and humanitarian disaster initiatives in the southern border of Syria and Jordan.

There is a need to estimate the number of refugee tents accurately to provide humanitarian needs and respond to appropriate camp-site planning, operations, and rescue effort. Managing a large refugee population in need of food, shelter, medicine, and other items to survive can be challenging.

Building on the existing satellite imaging such as CNN (convolutional neural network), CNN has been widely used in the remote-sensing research community, including urban planning, land use, and cover detection. Over the years, different climate and geographical regions present more challenging factors to gather accurate imaging and demands to perform for more sophisticated fine object detection. FCN provides the solution to magnify images and works well for multi-spectral satellite remote sensing and extracting images from crowded small objects. The FCN performs end-to-end, pixel-to-pixel inference for semantic segmentation in the images with arbitrary sizes.

For the IEEE Spectrum story, click here.



About Dr. Jiang Li

Dr. Jiang Li joined Old Dominion University in 2007 as an assistant professor in the Batten College of Engineering and Technology. Li received his Ph.D. degree in electrical engineering from the University of Texas at Arlington, TX, in 2004. In 2000, he received his master's degree in automation from Tsinghua University in China. In 1992 he received his bachelor of science in electrical engineering from Shanghai Jiaotong University, China. From 2004-06, Li worked as a postdoctoral fellow at the department of radiology of the National Institutes of Health.

Li recalled his interests in engineering began when his older brother came back from college one semester and told him how engineering is very interesting. Since Li lived in a rural village in China, information was very scarce back then. His brother bought books and magazines so Li could be more immersed with the subject. He also shared that from his early age, he dreamed of becoming a physicist since he was very good at math and physics. "I found that inventing new things that didn't exist before were actually more interesting to me, and later on I pursued engineering as my major in college," said Li.

Li's thesis and dissertation work focused on machine applied learning to image processing focused on medical image analysis. Over the years, his research interests shifted to applying deep learning techniques to remote sensing image analysis.