Author(s)

Qingfeng Tan, Anxi Yu, Zhihua He and Jiahao Tian

Corresponding Author

Anxi Yu

ABSTRACT

Video Synthetic Aperture Radar (SAR) is a land imaging mode where radar operates in spotlight mode for a long period of time. Its high frame rate imaging feature can provide the position of the target of interest in the spotlight scene for real-time detection and tracking. The shadow features of video SAR moving targets in the video sequence are difficult to be extracted using traditional methods such as frame difference method and background difference method. At the same time, due to the complex processing steps of traditional method, the detection threshold of targets is usually dynamically adjusted to detect different states of targets. Detection effectiveness and robustness are difficult to meet the requirements of moving target detection applications. However, due to the high number of feature extraction networks in deep convolutional networks, multi-scale shadow features of moving targets can be effectively extracted after convolution operations, and because the deep learning method converts the target detection problem into a regression problem, the detection speed much higher than traditional methods. In this paper, we propose a state-ofthe-art moving target detection method based on a YOLO framework. First, data augmentation is performed on the Sandia ViSAR data. At this stage, geometric transformation and color transformation methods are used. Second, in the stage of training the detection model, K-Means clustering is used to screen the sample target by region of interest, determine the YOLO anchor, and finally perform target detection on the testset sample, and the detected AP as an indicator is given for evaluating the detection method process, and do a comparison experiment with the traditional video SAR moving target detection method to verify the effectiveness of this method. The experimental results show that the video SAR moving target detection method based on the YOLO framework increases the detection probability by 28% compared with the traditional frame difference method.

KEYWORDS

Video synthetic aperture radar, moving target detection, deep convolutional networks, YOLO framework