“The Eagle” Remote Sensing Reconnaissance System

2018 Google - China MoE University-Industry Collaboration Program

This project intends to adopt a light-weight compression deep learning algorithm, based on Google open source artificial intelligence learning system TensorFlow, to propose and implement a set of intelligent processing solutions for mobile remote sensing images (target detection and semantic segmentation) on the mobile artificial intelligence computing platform.

The mobile platform can be carried on the satellite to realize on-orbit real-time remote sensing image information extraction, which provides a basis for subsequent data classification and efficient retrieval, and can filter the target data of interest according to user instructions for downlink transmission, thereby greatly reducing data. The transmission volume, alleviating the bandwidth pressure of the network link, reducing the energy consumption of data transmission, improving the targeting of the downlink data and the timeliness of information acquisition.

Video Preview:

Timeline and Outcomes

2018.05 Project Started
2018.06 Attending in 2018 Internet+ Innovation & Start-up Competition
2018.07 Won the Third-class award in 2018 Internet+ Innovation & Start-up Competition
2018.09 Attending in 2018 AI+ Innovation & Start-up Competition (report accessable on GitHub)
2018.10 Won the Third-Class Most Innovative Award in 2018 AI+ Innovation & Start-up Competition
2019.05 Won the National Second-Class Award in 2019 National College Student Innovation and Entrepreneurship Competition
2019.08.02 Oral presentation on IGARSS19
2019.11.14 Paper “Efficient Multi-Class Semantic Segmentation of High Resolution Aerial Imagery with Dilated LinkNet” was added to IEEE Xplore

Code on GitHub (managed by Qingtian Zhu)