A Method for Detecting SAR Vehicle Targets at Different Scales in Complex Background

Recently, deep convolutional neural networks have been widely used in target detection of synthetic aperture radar (SAR) images. However, the scenes in SAR images are often complex and varied, and the shapes and scales of targets vary greatly, which pose great difficulties for target detection. In this study, we propose a SAR image target detection method based on feature enhancement and feature balance. The approach begins by using deformable convolution to extract key features of the image more flexibly, expanding the range of receptive field. Subsequently, it utilizes the feature balance module to integrate the feature information of different levels and resolutions obtained by the neck network, thereby enhancing the network's ability to integrate target feature information in complex environments. The paper validates our proposed method on the self-developed FAST datasets. The experimental results show that compared with YOLOv7 and YOLOv8, which have strong detection performance, this method has better detection performance for complex scenes.