地球信息科学学报 ›› 2015, Vol. 17 ›› Issue (9): 1080-1091.doi: 10.3724/SP.J.1047.2015.01080

• • 上一篇    下一篇

高分辨率遥感影像目标分类与识别研究进展

刘扬1,2,3(), 付征叶4, 郑逢斌2,3,*()   

  1. 1. 河南大学环境与规划学院,开封 475004
    2. 河南大学空间信息处理实验室,开封 475004
    3. 河南大学计算机与信息工程学院,开封 475004
    4. 河南大学软件学院,开封 475004
  • 收稿日期:2014-12-12 修回日期:2015-02-14 出版日期:2015-09-10 发布日期:2015-09-07
  • 通讯作者: 郑逢斌 E-mail:ly.sci.art@gmail.com;zhengfb@henu.edu.cn
  • 作者简介:

    作者简介:刘 扬(1971-),男,河南信阳人,副教授,博士,主要从事媒体神经认知计算、时空信息高性能计算。E-mail: ly.sci.art@gmail.com

  • 基金资助:
    国家自然科学基金项目(61305042、61202098);国防科技工业民用专项科研技术研究项目(2012A03A0939);河南省教育厅科学技术研究重点项目(13A520071)

Review on High Resolution Remote Sensing Image Classification and Recognition

LIU Yang1,2,3(), FU Zhengye4, ZHENG Fengbin2,3,*()   

  1. 1. College of Environment and Planning, Henan University, Kaifeng 475004, China
    2. Laboratory of Spatial Information Processing, Henan University, Kaifeng 475004, China
    3. College of Computer Science and Information Engineering, Henan University, Kaifeng 475004, China
    4. College of Software, Henan University, Kaifeng 4750041, China
  • Received:2014-12-12 Revised:2015-02-14 Online:2015-09-10 Published:2015-09-07
  • Contact: ZHENG Fengbin E-mail:ly.sci.art@gmail.com;zhengfb@henu.edu.cn
  • About author:

    *The author: SHEN Jingwei, E-mail:jingweigis@163.com

摘要:

高分辨率遥感影像的目标分类与识别,是对地观测系统进行图像分析理解,以及自动目标识别系统提取目标信息的重要手段。本文综述了当前国内外在可见光、红外、合成孔径雷达和合成孔径声纳等遥感影像的目标分类与识别的关键技术和最新研究进展。首先,讨论了高分辨率遥感影像的目标分类与识别问题的主要研究层次和内容;其次,深入分析了高分辨率遥感影像目标分类与识别,在滤波降噪、特征提取、目标检测、场景分类、目标分类和目标识别的关键技术及其所存在的问题;最后,结合并行计算、神经计算和认知计算等技术,讨论了目标分类与识别的可行性方案。具体包括:(1)高性能并行计算在高分辨率遥感图像处理的主流技术,并给出了基于Hadoop+OpenMP+CUDA的高分辨率遥感影像混合并行处理架构;(2)深度学习对于提升目标分类和识别精度的应用前景,以及基于深度神经网络的多层次遥感影像目标识别方法;(3)认知计算在解决遥感影像大数据不确定性分析的模型与算法,并讨论了层次主题模型的多尺度遥感影像场景描述方案。此外,根据媒体神经认知计算的相关研究,探讨了遥感影像大数据的目标分类和识别的发展趋势和研究方向。

关键词: 目标分类与识别, 媒体神经认知计算, 并行计算, 深度学习, 主题模型

Abstract:

Target classification and recognition (TCR) of high resolution remote sensing image is an important approach of image analysis, for the understanding of earth observation system (EOS), and for extracting information from the automatic target recognition (ATR) system, which has important values in military and civil fields. This paper reviews the latest progress and key technologies between domestic and international remote sensing image TCR in optical, infrared, synthetic aperture radar (SAR) and synthetic aperture sonar (SAS). The main research levels and the contents of high resolution remote sensing image TCR are firstly discussed. Then, the key technologies and their existing problems of high resolution remote sensing image TCR are deeply analyzed, from aspects such as filtering and noise reduction, feature extraction, target detection, scene classification, target classification and target recognition. Finally, combined with the related technologies including parallel computing, neural computing and cognitive computing, the new methods of TCR are discussed. Specifically, the main framework includes three aspects, which are detailed in the following. Firstly, the predominant techniques of high resolution remote sensing image processing are discussed based on high performance parallel computing. And the hybrid parallel architecture of high resolution remote sensing image processing based on Apache Hadoop, open multi-processing (OpenMP) and compute unified device architecture (CUDA) are also presented in this paper. Secondly, application prospects of TCR accuracy promotion are analyzed based on a thorough study of neuromorphic computing, and the method of multi-level remote sensing image target recognition based on the deep neural network (DNN) is introduced. Thirdly, the model and algorithm of big data uncertainty analysis for remote sensing images are discussed based on probabilistic graphical model (PGM) of cognitive computing, and the multi-scale remote sensing image scene description is given based on hierarchical topic model (HTM). Moreover, according to the related research of multi-media neural cognitive computing (MNCC), we discuss the development trend and research direction of TCR for remote sensing images big data in the future.

Key words: target classification and recognition, multi-media neural and cognitive computing, parallel computing, deep learning, topic model