Author:
Zhou Qihui,Cui He,Liang Shengming,Li Huiqin
Abstract
Abstract
Aiming at addressing the contradiction between the information timeliness and on-board computing capacity and data transmission capacity in on-orbit image processing system of small commercial remote sensing satellites, a convolutional neural network based target detection algorithm is proposed. The calculation amount of the model is optimized to 25MFLOPS by using transfer learning, sparse training and weight quantification, while the weight file data is reduced from tens of Mb to 0.5Mb, which makes it possible for small commercial remote sensing satellite to detect target on-orbit quickly, and meets the requirements of the timeliness of remote sensing information.
Subject
General Physics and Astronomy
Reference16 articles.
1. Applications and development of ultra large aperture space optical remote sensors[J];Zhang;Optics and Precision Engineering,2016
2. Review on mapping space remote sensor optical system [J];Gong;Chinese Optics,2015
3. Optical Remote Sensing Technology Progress Promoted by ZY-1Series Satellites [J];Zhang;Spacecraft Recovery & Remote Sensing,2018
4. Applications of Deep Convolutional Neural Network in Computer Vision [J];Lu;Journal of Data Acquisition & Processing,2016
Cited by
1 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献