光学遥感图像的目标检测方法研究

Research on target detection method of optical remote sensing image

如何利用获取的高分辨率遥感图像进行精确的目标检测,是目前需要解决的一个重要问题。提出以一种基于改进YOLO算法的卷积神经网络,给出的模型采用CSPDrakNet为基本网络,并融合Foucs网络模块以进行光学遥感图像的目标检测,以提高目标检测精度。实验结果表明,提出网络的平均检测精度高达92.98%,比DrakNet53基本网络提高了8.55%,且检测时间低于YOLOv3,YOLOv4网络,具有更快的检测速度。

The use of optical remote sensing images for target detection has a wide range of applications.How to use the acquired high-resolution remote sensing images for accurate target detection is an important problem that needs to be solved at present.In this paper,we propose a convolutional neural network based on the improved YOLO(CNNIY)algorithm.The proposed model uses CSPDrakNet as the basic network and fuses Foucs network modules for target detection of optical remote sensing images.to improve the target detection accuracy.The experimental results show that the average detection accuracy of the proposed network is up to 92.98%,which is 8.55% better than the DrakNet53 basic network,and the detection time is lower than that of YOLOv3 and YOLOv4 networks,which has faster detection speed.