中国邮电高校学报(英文) ›› 2020, Vol. 27 ›› Issue (6): 87-100.doi: 10.19682/j.cnki.1005-8885.2020.0048

• Image Processing • 上一篇    

Parallel design of convolutional neural networks for remote sensing images object recognition based on data-driven array processor

山蕊1,3,蒋林2, 邓军勇1,3,崔朋飞1,3,张玉婷1,3,吴皓月1,3,谢晓燕1,3   

  1. 1. 西安邮电大学
    2. 西安科技大学
    3. 西安邮电大学
  • 收稿日期:2019-06-24 修回日期:2020-12-16 出版日期:2020-12-31 发布日期:2020-12-31
  • 通讯作者: 山蕊 E-mail:shanrui0112@163.com
  • 基金资助:
    数据驱动轻核阵列处理器自重构机制研究;效能驱动的光互连视频阵列处理器动态自重构体系结构;可编程动态自重构三维阵列芯片体系结构关键技术;三维光电混合片上网络关键技术研究;性能驱动可编程自重构图形处理器体系结构研究;陕西省科技统筹创新项目;陕西省重点研发计划

Parallel design of convolutional neural networks for remote sensing images object recognition based on data-driven array processor

Shan Rui, Jiang Lin, Deng Junyong, Cui Pengfei, Zhang Yuting, Wu Haoyue, Xie Xiaoyan   

  • Received:2019-06-24 Revised:2020-12-16 Online:2020-12-31 Published:2020-12-31
  • Contact: SHAN Rui E-mail:shanrui0112@163.com
  • Supported by:
    Research on Self-reconfiguration Mechanism of Data Driven Light Core Array Processor;Performance-driven optical interconnect video array processor dynamic self-reconfiguration architecture;Key Technology of Programmable Dynamic Self-Reconfigurable 3D Array Chip Architecture;Research on Key Technologies of 3D Photoelectric Hybrid Network;Performance-Driven Programmable Self-Reconfigurable Graphics Processor Architecture;the Shaanxi Provincial Co-ordination Innovation Project of Science and Technology;the Shaanxi Provincial key R & D plan

摘要:

Object recognition in very high-resolution remote sensing images is a basic problem in the field of aerial and
satellite image analysis. With the development of sensor technology and aerospace remote sensing technology, the quality and quantity of remote sensing images are improved. Traditional recognition methods have a certain limitation in describing higher-level features, but object recognition method based on convolutional neural network (CNN) can not only deal with large scale images, but also train features automatically with high efficiency. It is mainly used on object recognition for remote sensing images. In this paper, an AlexNet CNN model is trained using 2 100 remote sensing images, and correction rate can reach 97.6% after 2 000 iterations. Then based on trained model, a parallel design of CNN for remote sensing images object recognition based on data-driven array processor (DDAP) is proposed. The consuming cycles are counted. Simultaneously, the proposed architecture is realized on Xilinx V6 development board, and synthesized based on SMIC 130 nm complementary metal oxid semiconductor (CMOS) technology. The experimental results show that the proposed architecture has a certain degree of parallelism to achieve the purpose of accelerating calculations.

关键词: convolutional neural networks, remote sensing images, object recognition, array processor, data-driven

Abstract:

Object recognition in very high-resolution remote sensing images is a basic problem in the field of aerial and
satellite image analysis. With the development of sensor technology and aerospace remote sensing technology, the quality and quantity of remote sensing images are improved. Traditional recognition methods have a certain limitation in describing higher-level features, but object recognition method based on convolutional neural network (CNN) can not only deal with large scale images, but also train features automatically with high efficiency. It is mainly used on object recognition for remote sensing images. In this paper, an AlexNet CNN model is trained using 2 100 remote sensing images, and correction rate can reach 97.6% after 2 000 iterations. Then based on trained model, a parallel design of CNN for remote sensing images object recognition based on data-driven array processor (DDAP) is proposed. The consuming cycles are counted. Simultaneously, the proposed architecture is realized on Xilinx V6 development board, and synthesized based on SMIC 130 nm complementary metal oxid semiconductor (CMOS) technology. The experimental results show that the proposed architecture has a certain degree of parallelism to achieve the purpose of accelerating calculations.

Key words: convolutional neural networks, remote sensing images, object recognition, array processor, data-driven

中图分类号: