计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (18): 209-213.DOI: 10.3778/j.issn.1002-8331.2004-0410

• 图形图像处理 • 上一篇    下一篇

基于改进Yolo v3算法的遥感建筑物检测研究

董彪,熊风光,韩燮,况立群,徐清宇   

  1. 1.中北大学 大数据学院,太原 030051
    2.北方自动控制技术研究所 仿真装备研发部,太原 030051
  • 出版日期:2020-09-15 发布日期:2020-09-10

Research on Remote Sensing Building Detection Based on Improved Yolo v3 Algorithm

DONG Biao, XIONG Fengguang, HAN Xie, KUANG Liqun, XU Qingyu   

  1. 1.College of Big Data, North University of China, Taiyuan 030051, China
    2.Simulation Equipment R&D Department, North Automatic Control Technology Institute, Taiyuan 030051, China
  • Online:2020-09-15 Published:2020-09-10

摘要:

针对遥感图像中建筑物检测存在小型建筑物检测难度大、检测过程中无法满足实时性等问题,提出将基于深度学习的目标检测算法Yolo v3应用到建筑物检测场景中。以实时性及泛用性良好的Yolo v3为基本算法,满足实时性的要求;通过改进Yolo v3的网络结构,以修改特征图分辨率、调整先验框维度为方向加强对小型建筑物的检测能力。实验结果表明,改进的Yolo v3目标检测算法既满足了实时性的要求,且检测精度和召回率达到了91.29%和95.61%,较原算法分别提高了5.35%和2.34%。因此提出的改进方法有效解决了遥感领域小型建筑物的检测问题。

关键词: 目标检测, 遥感, 深度学习, Yolo v3, 图像处理

Abstract:

In order to solve the problems in building detection for remote sensing images, such as the difficulty of detecting small buildings and the lack of real-time performance in the detection process, the deep learn-based target detection algorithm Yolo v3 is applied to the building detection scene. The basic algorithm of Yolo v3 with good real-time and universality is adopted to meet the real-time requirements. By improving the network structure of Yolo v3, the detection ability of small buildings is strengthened by modifying the resolution of feature map and adjusting the dimension of prior frame. The experimental results show that the improved Yolo v3 target detection method meets the real-time requirements and the detection accuracy and recall rate reaches 91.29% and 95.61%, which are 5.35% and 2.34% higher than the original algorithm. Therefore, the improved method effectively solves the detection problem of small buildings in remote sensing field.

Key words: target detection, remote sensing, deep learning, Yolo v3, image processing