地球信息科学学报 ›› 2018, Vol. 20 ›› Issue (10): 1500-1508.doi: 10.12082/dqxxkx.2018.180237.

• 遥感科学与应用技术 • 上一篇    下一篇

基于Faster R-CNN深度网络的遥感影像目标识别 方法研究

王金传1(), 谭喜成2,3,*(), 王召海1, 钟燕飞4, 董华萍4,5, 周松涛2, 成布怡2   

  1. 1. 山东师范大学地理与环境学院,济南 250358
    2. 武汉大学遥感信息工程学院,武汉 430079
    3. 地理空间信息与数字技术国家测绘地理信息局工程技术研究中心,武汉 430079
    4. 武汉大学 测绘遥感信息工程国家重点实验室,武汉 430079
    5. 城市空间信息工程北京市重点实验室,北京 100038
  • 收稿日期:2018-05-12 修回日期:2018-07-25 出版日期:2018-10-25 发布日期:2018-10-17
  • 通讯作者: 谭喜成 E-mail:wangjnchuan@163.com;xctan@whu.edu.cn
  • 作者简介:

    作者简介:王金传(1993-),男,硕士,主要从事遥感技术应用、深度学习和WebGIS应用研究。E-mail: wangjnchuan@163.com

  • 基金资助:
    国家重点研发计划项目(2017YFB0504202);国家自然科学基金项目(41871312);湖北省自然科学基金项目(2017CFB433);智能地学信息处理湖北省重点实验室(中国地质大学(武汉))开放基金(KLIGIP-2017A09);上海航天科技创新基金项目(SAST2016006);空间数据挖掘与信息共享教育部重点实验室(福州大学)开放基金(2016LSDMIS06);城市空间信息工程北京市重点实验室经费资助项目(2017209)

Faster R-CNN Deep Learning Network Based Object Recognition of Remote Sensing Image

WANG Jinchuan1(), TAN Xicheng2,3,*(), WANG Zhaohai1, ZHONG Yanfei4, DONG Huaping4,5, ZHOU Songtao2, CHENG Buyi2   

  1. 1. College of Geography and Environment, Shandong Normal University, Jinan 250358, China
    2. School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China
    3. Engineering Research Center of Geospatial Information and Digital Technology, NASG, Wuhan 430079, China
    4. State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China
    5. Beijing Key Laboratory of Urban Spatial Information Engineering, Beijing 100038, China
  • Received:2018-05-12 Revised:2018-07-25 Online:2018-10-25 Published:2018-10-17
  • Contact: TAN Xicheng E-mail:wangjnchuan@163.com;xctan@whu.edu.cn
  • Supported by:
    National Key Research and Development Program of China under Grant, No.2017YFB0504202;National Natural Science Foundation of China, No.41871312;Hubei Natural Science Foundation, No.2017CFB433;Hubei Key Laboratory of Intelligent Geo-Information Processing (China University of Geosciences (Wuhan)), No. KLIGIP-2017A09;The Key Laboratory of Spatial Data Mining & Information Sharing of the Ministry of Education, Fuzhou University, No.2016LSDMIS06;Shanghai Aerospace Science and Technology Innovation Fund, No.SAST2016006;Beijing Key Laboratory of Urban Spatial Information Engineering, No.2017209.

摘要:

遥感影像目标识别在众多领域中具有极高的理论意义与应用价值,更快速、更精确的目标识别方法研究是目前遥感及图像研究领域的热点与难点。本文将深度学习的方法应用于遥感影像目标识别中,提出基于Faster R-CNN深度学习网络的目标快速精确识别方法。该方法采用了包括基于RPN的建议区域提取方法和VGG16训练卷积网络模型,构建了面向遥感影像目标识别的深度卷积神经网络。为验证该方法的精度及性能,在Caffe深度学习框架上,选取高分辨率遥感影像中飞机、油罐、操场及立交桥目标进行验证实验。结果表明,基于Faster R-CNN的深度学习方法能够实现对遥感影像目标的快速、准确识别,同时具有较好的推广性。通过本文的研究,证明基于Faster R-CNN深度学习的高分遥感影像目标识别方法具有显著优势和潜力,对基于其他深度学习方法的目标识别研究也有一定的参考意义。

关键词: 深度学习, Faster R-CNN网络, 高分辨率, 遥感影像, 目标识别

Abstract:

Object recognition of remote sensing image is of great theoretical significance and application value in many fields. Faster and more accurate object identification methods are hot and difficult points in the field of remote sensing and image. In this paper, the method of deep learning is applied to remote sensing image object recognition, and a fast and accurate method of object recognition based on Faster R-CNN deep learning network is proposed. This method uses the proposal region extraction method based on RPN and the VGG16 training convolution network model, and constructs a deep convolutional neural network for the object recognition of remote sensing image. In order to verify the accuracy and performance of the method, the GPU accelerated computing model was used in the Caffe deep learning framework. Firstly, the aircraft target recognition experiment in remote sensing image was designed. The aircraft target recognition accuracy reached 96.67%. Then, after the experiment was successful, we continued to identify other target features, and selected the high-resolution remote sensing image of the oil tank, playground and overpass object for verification experiments. In the same experiment environment, the same good experimental verification results were obtained, the target recognition rate was at a high level, and the cost time of recognition in each picture was less than 0.2 seconds, which fully verified the validity and reliability of the model studied in this paper. After analysis and comparison, the conclusion is that the deep learning method based on Faster R-CNN can realize the fast and accurate recognition of the selected targets, which proves that the method has a good promotion significance in high-resolution remote sensing image target recognition applications. Therefore, the model has great application value, and it also has certain reference significance for target recognition research based on other deep learning methods.

Key words: deep learning, Faster R-CNN, high resolution, remote sensing image, object recognition