Moving Vehicle Detection Based on Millimeter Wave ISAR Image with Range-Doppler Map

In the target detection task in synthetic aperture radar (SAR) imagery, the target may become defocused due to its own motion, leading to detection failure. To address the issue of SAR motion-defocused target detection, this paper first uti-lizes the millimeter-wave radar inverse SAR (ISAR) imaging method to collect and construct an ISAR moving vehicle detection dataset (IMVDD), then proposes a range-Doppler (RD) map guided detection network (RDM-Net). The ISAR images demonstrate that vehicle targets are azimuthally de-focused and significantly affected by clutter, presenting as a composition of discrete strong scatterers. Based on YOLOv8 model, an azimuthal spatial attention constructed from range-Doppler map is introduced to utilize the implicit motion information in SAR data to aid the localization of targets. Moreover, the coordinate attention (CA) is introduced to better extract detailed and context information from SAR scattering features; then a modified asymptotic feature pyramid network (AFPN) is introduced to more effectively facilitate fusion the feature maps of different layers. Experiments conducted on the proposed IMVDD dataset demonstrate that the proposed method can effectively enhance defocused target detection performance.