一种基于场景合成和锚点约束的SAR目标检测网络

随着深度学习方法在计算机视觉领域的崛起，如何将其应用于具有全天时、全天候等优点的SAR图像也成为一大研究重点.相较于传统图像，SAR图像由于其难判读、应用人群较少等原因难以获得大量标注数据.本文提出一种基于场景合成和锚点约束的SAR图标检测方法.通过区域生长算法和阈值法对SAR车辆目标及其阴影进行分割，然后随机嵌入SAR复杂场景中的合理区域来合成目标检测数据集.针对SAR车辆目标的几何特性、图像分辨率参数，对Faster-RCNN中的锚点大小进行约束，减少不符合SAR车辆目标检测框尺寸的候选框，大量约简冗余计算，提升训练、测试效率及精度.

SAR target detection network based on scenario synthesis and anchor constraint

Deep learning methods such as convolutional neural networks (CNN) have been widely used in fields of image processing and object recognition.However,the SAR images cannot yet be efficiently detected by CNN methods.Compared with traditional images,the SAR images have the advantage of all-day and all-weather acquisition,but they cannot obtain enough annotation due to the difficulty for interpretation and short of users.This paper proposes a SAR target detection method based on scene synthesis and anchor constraint.Firstly,the target and its shadow are segmented by region growing as well as threshold algorithm,and then the target detection data set is synthesized by randomly embedding the reasonable region into the SAR complex scene.Considering the SAR target''s geometric characteristics and image resolution parameters,the anchor''s size of Faster-RCNN is constrained to reduce the candidate frames that cannot meet the SAR target detection frame size,which massively reduce redundancy calculations so as to improve the efficiency and accuracy of training and testing process.