Please wait a minute...
img

官方微信

遥感技术与应用  2018, Vol. 33 Issue (6): 1095-1102    DOI: 10.11873/j.issn.1004-0323.2018.6.1095
数据与图像处理     
基于深度学习的国产高分遥感影像飞机目标自动检测
李淑敏1,冯权泷2,梁其椿1,张学庆1
(1.中电科海洋信息技术研究院有限公司,北京 100041;
2.北京航天宏图信息技术股份有限公司,北京 00195)  
 Aircraft Auto-detection in Domestic High Resolution Remote SensingImages Using Deep-learning
Li Shumin1,Feng Quanlong2,Liang Qichun1,Zhang Xueqing1
(1.CETC Ocean Information Technology Co.,Ltd.,Beijing 100041,China;
2.BeijingPiesat Information Technology Co.,Ltd.,Beijing 100195,China)

 全文: PDF(7505 KB)  
摘要:

基于高分辨率遥感影像进行飞机目标的自动检测对精确掌握飞机的空间位置具有重要意义。由于飞机姿态不一、背景复杂、轮廓不完整等原因,导致飞机自动检测的难度较大、检测精度不高。传统飞机检测方法主要基于人工特征和机器学习分类器,在算法鲁棒性、位移、旋转不变性等方面表现欠佳。为了解决上述问题,通过引入深度神经网络模型和迁移学习策略,基于国产高分辨率遥感影像实现了飞机目标的高精度检测。具体而言,首先构建了多尺度飞机样本数据库,并基于YOLO V2目标检测模型进行迁移学习,从而提高飞机检测模型的精度和鲁棒性。以上海浦东机场GF-影像为例进行飞机目标检测实验,实验结果表明:飞机召回率为92.25%,正确率为94.93%;通过深度学习和模型迁移可以实现小样本情况下的飞机目标高精度检测。该方法可以推广到其他地物的检测和识别中,具有较好的推广意义和价值。

关键词: 遥感影像深度学习目标检测飞机卷积神经网络    
Abstract: It is of great significance to automatically detect aircrafts from remote sensing imagery to get their locations.However,due to aircraft posture variance,complicated background and incomplete outlines,it is challenging to achieve a high aircraft detection accuracy.Traditional aircraft detection methods are usually based on hand\|crafted features and machine learning based classifiers,which is not robust enough for the translation and rotation variations.To tackle the above issues,this paper introduces deep convolutional neural network and the strategy of transfer learning to detect aircrafts from Chinses domestic satellite remote sensing images.Specifically,this paper first constructs an aircraft sample database,which consists aircrafts of different sizes and poses.Afterwards,YOLO V2 trained with natural images is utilized as the detection model and is further fine\|tuned with aircraft samples to increase the robustness and performance.Experiments were done on the Shanghai Pudong airport from Chinese GF\|2 remote sensing data.Experimental results showed a good performance with a recall of 92.25% and a precision of 94.93%.It is indicated that deep learning together with model transfer can get a high aircraft detection accuracy with limited training samples.The method in this paper can be generalized to other land object detection problems which shows a good promotional value.
Key words: Remote sensing image    Deep learning    Object detection    Aircraft    Convolution neural network
收稿日期: 2017-10-24 出版日期: 2019-01-29
ZTFLH:  TP391.4  
作者简介: 李淑敏(1983- ),女,河北沧州人,高级工程师,主要从事资源、环境、海洋等遥感应用技术研究。Email:lism@cetcocean.com。
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章  

引用本文:

李淑敏, 冯权泷, 梁其椿, 张学庆. 基于深度学习的国产高分遥感影像飞机目标自动检测[J]. 遥感技术与应用, 2018, 33(6): 1095-1102.

Li Shumin, Feng Quanlong, Liang Qichun, Zhang Xueqing.  Aircraft Auto-detection in Domestic High Resolution Remote SensingImages Using Deep-learning. Remote Sensing Technology and Application, 2018, 33(6): 1095-1102.

链接本文:

http://www.rsta.ac.cn/CN/10.11873/j.issn.1004-0323.2018.6.1095        http://www.rsta.ac.cn/CN/Y2018/V33/I6/1095

[1] 冀新莹, 韦玉春, 王问尧, 方宏. 城镇区域高分辨率遥感影像地表覆盖变化检测的误差分析[J]. 遥感技术与应用, 2018, 33(5): 932-941.
[2] 周子勇. 运用经验模态分解和压缩感知方法进行遥感影像超分辨率重建[J]. 遥感技术与应用, 2018, 33(1): 96-102.
[3] 周为峰,曹利,李小恕,程田飞. 沿海牡蛎养殖的WorldView-2影像融合方法评价[J]. 遥感技术与应用, 2018, 33(1): 103-109.
[4] 熊伟,徐永力,姚力波,崔亚奇. 基于SVM的高分辨率SAR图像舰船目标检测算法[J]. 遥感技术与应用, 2018, 33(1): 119-127.
[5] 田德宇,张耀南,赵国辉,韩立钦. 基于卷积神经网络的遥感沙漠绿地提取方法[J]. 遥感技术与应用, 2018, 33(1): 151-157.
[6] 何海清,庞燕,陈晓勇. 面向遥感影像场景的深度卷积神经网络递归识别模型[J]. 遥感技术与应用, 2017, 32(6): 1078-1082.
[7] 林齐根,邹振华,祝瑛琦,王瑛. 基于光谱、空间和形态特征的面向对象滑坡识别[J]. 遥感技术与应用, 2017, 32(5): 931-937.
[8] 吕利利,颉耀文,董龙龙. 基于不同地形校正模型的影像反射率对比分析[J]. 遥感技术与应用, 2017, 32(4): 751-759.
[9] 吕利利,颉耀文,黄晓君,张秀霞,李汝嫣. 基于CART决策树分类的沙漠化信息提取方法研究[J]. 遥感技术与应用, 2017, 32(3): 499-506.
[10] 王艳楠,王健健,龚健新,袁帅,刘辉,罗文. 基于环境卫星数据的沿海滩涂地物类型分类的随机森林方法[J]. 遥感技术与应用, 2016, 31(6): 1107-1113.
[11] 朱济帅,尹作霞,谭琨,王雪,李二珠,杜培军. 基于空间邻域信息的高光谱遥感影像半监督协同训练[J]. 遥感技术与应用, 2016, 31(6): 1122-1130.
[12] 吴阿丹,郭建文. 遥感影像Web共享关键技术应用研究[J]. 遥感技术与应用, 2016, 31(6): 1209-1214.
[13] 李盛阳,于海军,韩洁,黑保琴. 基于三维地球的海量遥感影像高效可视化管理系统的设计与实现[J]. 遥感技术与应用, 2016, 31(1): 170-176.
[14] 毛召武,程结海,袁占良. 顾及面积和位置差异的高分遥感影像分割质量评价方法[J]. 遥感技术与应用, 2016, 31(1): 186-193.
[15] 张文博,覃志豪,刘含海. 基于分形理论的土地利用类型变化研究—以广州番禺为例[J]. 遥感技术与应用, 2015, 30(5): 952-958.