A Novel Scene-Aware Pipeline Architecture for Spaceborne SAR Regional Imaging and Ship Target Detection

Abstract

With the rapid advancement of synthetic aperture radar (SAR) technology and the explosive growth of on-orbit data, conventional processing frameworks are increasingly constrained by high resource consumption and substantial latency, rendering them inadequate for real-time maritime ship detection on resource-limited platforms. To address this challenge, this article proposes a novel scene-aware imaging-detection integration (SAIDI) architecture for stripmap-mode SAR, which unifies traditionally decoupled imaging and detection tasks into a cascaded processing pipeline. Specifically, the framework comprises four key components: an imaging necessity judgment module to filter target-free raw echoes, a fast imaging algorithm bypassing range cell migration correction, a lightweight pruned YOLOv11 detection network, and a dedicated ship dataset constructed from real satellite measurements. Experiments on real satellite data demonstrate that the proposed method maintains robust detection performance in both open-ocean and coastal scenarios while eliminating redundant computations in background regions. Compared to conventional methods, the proposed scheme significantly reduces processing latency and peak memory footprint, while optimizing CPU utilization and power efficiency. In addition, this method is mainly designed for medium- and large-sized marine ships.