摘要
针对高分辨率遥感影像场景复杂,现有的目标识别算法检测率较低且速度较慢的问题,提出了一种改进的RFBNet模型。算法在RFBNet模型的基础上构建特征金字塔网络,融合高层语义信息和低层特征信息,提高了网络识别能力。为验证该算法性能,以遥感影像中飞机目标为例进行了实验验证。以油罐和立交桥目标为例,对该算法的推广性进行了实验验证。结果表明,改进的RFBNet模型在遥感影像目标识别中精度较高、速度较快,且具有较好的推广性。
In the remote sensing image, the background is complex, which results in the low detection rate and slow speed of some target recognition algorithms. Therefore, an improved RFB Net model is proposed. In the algorithm a feature pyramid network is built based on RFB Net model, high-level semantic information and low-level feature information are integrated to improve the network recognition ability. In order to verify the performance of the algorithm, the verification experiment is carried out by taking the aircraft target in the remote sensing image as an example. In addition, taking the tank and overpass target as an example, the generalization of the algorithm is validated. The results show that the improved RFB Net model has high precision and fast speed in remote sensing image target recognition, and has better generalization.
作者
刘相云
郭呈渊
龚志辉
金飞
余东行
LIU Xiangyun;GUO Chengyuan;GONG Zhihui;JIN Fei;YU Donghang(Information Engineering University, Zhengzhou 450001, China;Joint Servicing Academy of National Defence University, Beijing 100039, China;96901 Troops, Beijing 100039, China)
出处
《测绘科学技术学报》
北大核心
2019年第2期179-184,共6页
Journal of Geomatics Science and Technology
基金
信息工程大学校立项目(2016609602)
关键词
RFB网络
特征金字塔
遥感影像
飞机目标识别
深度学习
RFB Net
feature pyramid network
remote sensing image
aircraft targets recognition
deep learning
作者简介
刘相云(1994-),男,河南周口人,硕士生,主要研究方向为模式识别与机器学习。E-mail:liu_xy1994@163.com.
