基于深度学习的复杂沙漠背景SAR目标检测

SAR Target Detection in Complex Desert Background Images Based on Deep Learning

SAR目标检测,因成像场景大、背景复杂多变而极具挑战。传统基于恒虚警率的SAR目标检测方法极易受背景干扰。针对上述问题,提出一种基于深度学习的复杂沙漠背景SAR目标端对端检测识别系统。即采用小规模沙漠背景下的SAR图像数据对Faster-RCNN网络进行迁移训练,一体化完成典型目标的检测与识别。基于合成数据集Desert-SAR的试验结果表明,与传统方法相比,该方法检测速度更快、准确率更高、鲁棒性更强。

Target detection in synthetic aperture radar(SAR)image is a challenge due to the large-scale and complex imaging scene.The classical methods based on CFAR are sensible to imaging scene.Aiming at this problem,we propose an end-to-end target detection method for SAR image in desert scene based on deep learning.That is,the transfer learning is employed to adjust the Faster-RCNN network for optical image to the SAR image.Experimental results of the Dessert-SAR data set show that the proposed method can achieve faster detection speed,higher accuracy and robustness compared with the classical ones.