【作者】 王建；

【导师】 卢一相；

【作者基本信息】 安徽大学 ， 模式识别与智能系统， 2020， 硕士

【摘要】 雷达因其不受雾、云、雨等条件的阻挡,可以全天候、全天时探测远距离目标。雷达成为军事上必不可少的电子装备,也被广泛应用于资源探测、环境检测、海警执法和其他科学研究,例如:大气物理等。雷达回波图像是由反射回来的电磁波信号经过采样得到,由于没有经过滤波处理,图像中存在大量噪声,采用肉眼分辨具有一定的挑战性。因此,研究雷达回波图像的目标检测方法对于识别和快速做出决策具有重要意义。论文以雷达回波图像为研究对象,研究在不同工况下生成得到的回波图像目标检测性能。论文的主要工作如下:(1)研究了基本的滤波算法包括中值滤波、均值滤波、NL-mean(Non Local means)算法以及BM3D(Block-matching 3D filtering)去噪算法,并将其应用于雷达回波图像的去噪。实验结果表明,基于BM3D的图像去噪算法可以很好的过滤雷达回波图像上的噪声信号。同时,研究了BM3D算法中硬阈值对去噪效果的影响。(2)详细介绍了边缘检测的一般方法,包括Sobel算子、高斯拉普拉斯算子、Canny算子等基本理论。针对雷达回波图像,采用边缘检测算子获取图像中的目标信息。实验结果显示,在同一工况下,基于Canny算子的雷达回波图像目标检测虽然优于实验中的其他检测算子,但是总体的准确率不高。另外实验发现,针对雷达在不同工况时得到的雷达回波图像,在不改变Canny算子提取目标轮廓信息所用的阈值的情况下,检测的效果在准确性、可靠性上较差。得出的结论是:将Canny算子作为目标检测的算法镶嵌到雷达探测系统中不足以辅助雷达进行很好的避碰。(3)提出了基于轮廓提取的雷达回波图像目标检测方法,该方法的主要贡献有三点:第一,获取图像中所有的轮廓信息,包括真目标与假目标。第二,借鉴目标检测中IOU(Intersection Over Union,IOU)的概念,设定阈值比较两个目标轮廓信息之间的大小,过滤不符合的目标信息。第三,在同一工况下得到的雷达回波图像,能够检测到图像中的绝大多数目标,漏检误检相对于Canny算子较低,在不同工况下得到的雷达回波图像,也能具有很好的检测效果。实验表明,本章提出的方法对雷达回波图像的目标检测具有较好的准确性,同时具有一定的鲁棒性。更多还原

【Abstract】 Radar can detect long distance targets all day because it is not blocked by fog,clouds,rain and other conditions,so it becomes an indispensable electronic equipment in the military.It is also widely used in resource detection,environmental detection,marine police law enforcement and other scientific research,such as atmospheric physics.Radar echo image is obtained by sampling the reflected electromagnetic wave signal.Because there is a lot of noise in the image without filtering,it is challenging to use naked eye resolution.Radar echo image is obtained by sampling the reflected electromagnetic wave signal.Without filtering processing there is a lot of noise in the image,so it is challenging to use the naked eye to distinguish.Therefore,it is of great significance to study the target detection method of radar echo image for recognition and make quick decisions.In this paper,the radar echo image is taken as the research object,to study the target detection performance of the echo image generated under different working conditions.The main work of the paper is as follows:(1)Based on the characteristics of radar echo image with strong noise,the basic filtering algorithm is studied to improve the quality of radar echo image,including median filtering,mean filtering,NL-means(Non Local means)algorithm,BM3D(Block-matching 3D filtering)denoising algorithm.The experimental results shows that the image denoising algorithm based on BM3 D can make good performance on the radar echo image well,At the same time,we also do some researches on the influence of the hard threshold operator on the denoising effect in the BM3 D algorithm.(2)According to the characteristics of radar echo image itself,this paper introduces the processing effect of different image filtering algorithms for a large number of noises in radar echo image.Secondly,some algorithms for extracting the contour information of image are briefly introduced.The method of extracting target contour information with Canny operator as the core is used to obtain target information on radar echo image and then detect it.Finally,through the discussion and analysis of the experiment,the simple use of the Canny operator can not detect all the targets in the radar echo image,and the miss detection rate is relatively large.Besides,the experimental results showed that the detection effect was not satisfactory without modifying the threshold of extracting target contour information with Canny operator as the core.(3)This paper proposes a radar echo image target detection method based on contour extraction.There are two main contributions of this method: first,drawing on the concept of IOU(Intersection Over Union)in target detection,setting a threshold to compare the size between the two target contour information,and filtering the target information that does not meet the requirements.Second,the radar echo image obtained under the same working condition,can detect most of the targets in the image,error detection is lower than the Canny operator,and the radar echo image obtained under different working conditions can also have a good detection effect.Experiments show that the method proposed in this paper has good accuracy and robustness for target detection of radar echo images.更多还原