结合卷积神经网络与显著性特征的机场检测  

Airport detection using convolutional neural network and salient feature

作　　者：余东行 张宁 张保明[1] 郭海涛[1] 卢俊[1] YU Donghang;ZHANG Ning;ZHANG Baoming;GUO Haitao;LU Jun(Information Engineering University, Zhengzhou 450001, China;China National Administration of GNSS and Application, Beijing 100088, China)

机构地区：[1]信息工程大学,河南郑州450001 [2]中国卫星导航定位应用管理中心,北京100088

出　　处：《测绘通报》2019年第7期44-49,共6页Bulletin of Surveying and Mapping

基　　金：国家自然科学基金(41601507)

摘　　要：遥感影像机场检测中,针对传统人工设计特征的方法稳健性差、检测耗时的问题,提出了一种结合卷积神经网络与显著性特征的遥感影像机场检测算法。利用卷积神经网络快速准确地检测出机场目标,确定兴趣区域,对兴趣区域进行显著性检测和连通区提取,从而获取更加精确的机场边界,最后利用多种场景下的影像进行测试。结果表明,本文方法具有明显的精度和速度优势;利用频率视觉显著性分析方法对获得的机场区域进行视觉显著性检测,可有效获取机场和跑道的精确边界,提高机场检测的效果和实用价值。Existing algorithms of airport detection using handcraft features perform time-consuming and poor robustness. In view of these problems, this paper proposes a method using convolutional neural network and salient feature. First, a deep convolutional neural network is used to extract the regions of interest (ROI) from complex remote sensing images. Then, saliency detection based on frequency-tuned is introduced to get saliency map of those regions. Through segment on the saliency map and marking the connected region on the binary image, the maximum connected region which is most likely be area of the airport is extracted. Different kinds of airports are used to test and the results show that the proposed method has obvious advantages in precision and speed. With the aid of saliency detection, the precise boundary of the airport and runway can be obtained effectively and the effect and practical value of the airport detection are hugely improved.

关 键 词：机场检测 遥感影像 卷积神经网络 显著性特征 

分 类 号：P237[天文地球—摄影测量与遥感]