Research On Target Detection Of Low-Altitude Surveillance Radar Based On Deep Learning

In the field of low-altitude surveillance,the airborne equipment of small aircraft is re-stricted by power and cost,and it's difficult to broadcast the motion parameters of the aircraft accurately,resulting in it only can be actively detected by radar.This type of small aircraft has the characteristics of low flying height,slow flying speed and small radar reflection area,which greatly increases the difficulty of radar target detection.An increasing number of small aircraft have caused many air traffic control accidents.Therefore,improving the radar's ability to detect small targets in low-altitude airspace has become an urgent problem need to be solved in the air traffic control field in recent years.This paper has carried out in-depth research on radar target detection algorithms based on deep learning.According to the characteristics of low-altitude targets,we combined the radar signal processing technology and the target detection algorithm which based on deep learning,with the help of the echo data of a company's low-altitude surveillance radar,this paper proposed an improved moving target detection image and target detection algorithm optimization scheme to achieve the purpose of improving the target detection effect of low-altitude surveillance radar.The research content of this article mainly includes the following aspects:1.Proposed a low-distortion data set construction scheme based on moving target detec-tion.Based on the real echo data and its signal parameters,this paper designed a real echo decoding program and an analog echo generation program,and obtains a moving target detec-tion image through radar signal processing.In order to maximize the use of image information,this paper proposed that the real part,imaginary part,and modulus of the complex number ma-trix correspond to the three color channels of the image,and normalized storage is performed,which significantly increases the amount of information of a single pixel.Experiments that the low-distortion data set construction scheme proposed in this paper can improve the detection effect of each algorithm by more than 11%.2.Proposed a set of improvement schemes for low-altitude surveillance radar target de-tection algorithm based on deep learning.In order to further improve the detection effect of the target detection algorithm based on deep learning on the low-altitude surveillance radar target,the improvement points proposed in this paper are mainly:a.Correct the size of the initial an?chor point to speed up the regression speed of the preselection box;b.Use the Mish activation function Carry out gradient transfer;c.Introduce the idea of CSPNet to separate operations and speed up the inference speed;d.Design a new IOU calculation method to perform pre-selection box regression more efficiently.The above improvements have certain reference value for all radar target detection algorithms based on deep learning.3.Based on YOLOv3,Faster R-CNN and improved schemes,two deep learning-based low-altitude surveillance radar target detection algorithms,YOLOnew and Fasternew,are pro-posed.Perform data expansion and data enhancement when inputting data,and test its per-formance through four experiments.Based on the experimental results,after using the low-distortion matrix data set and model improvement scheme,compared with YOLOv3,the YOLOnew proposed in this paper has a 20.3%and 32.1%performance improvement on the simulated data set and the real data set,respectively.Compared with Faster R-CNN,the Fasternew proposed in this paper has a performance improvement of 12.9%and 5.3%on the simulated data set and the real data set,respectively.Among them,the average detection time of YOLOnew is 20ms,which meets the requirements of high real-time ability.According to the experimental results,in terms of common technology,the low-distortion data set construction method and target detection algorithm improvement scheme proposed in this paper have a certain universality and can be applied to more target detection algorithms based on deep learning.In terms of engineering projects,YOLOnew can be used in existing low-altitude surveillance radar target detection projects due to its excellent real-time capability and high accuracy.